smftools 0.1.7__py3-none-any.whl → 0.2.1__py3-none-any.whl

smftools/config/experiment_config.py

@@ -0,0 +1,1191 @@
+# experiment_config.py
+from __future__ import annotations
+import ast
+import json
+import warnings
+from dataclasses import dataclass, field, asdict
+from pathlib import Path
+from typing import Any, Dict, List, Optional, Tuple, Union, IO, Sequence
+
+# Optional dependency for YAML handling
+try:
+    import yaml
+except Exception:
+    yaml = None
+
+import pandas as pd
+import numpy as np
+
+
+# -------------------------
+# Utility parsing functions
+# -------------------------
+def _parse_bool(v: Any) -> bool:
+    if isinstance(v, bool):
+        return v
+    if v is None:
+        return False
+    s = str(v).strip().lower()
+    if s in ("1", "true", "t", "yes", "y", "on"):
+        return True
+    if s in ("0", "false", "f", "no", "n", "off", ""):
+        return False
+    try:
+        return float(s) != 0.0
+    except Exception:
+        return False
+
+
+def _parse_list(v: Any) -> List:
+    if v is None:
+        return []
+    if isinstance(v, (list, tuple)):
+        return list(v)
+    s = str(v).strip()
+    if s == "" or s.lower() == "none":
+        return []
+    # try JSON
+    try:
+        parsed = json.loads(s)
+        if isinstance(parsed, list):
+            return parsed
+    except Exception:
+        pass
+    # try python literal eval
+    try:
+        lit = ast.literal_eval(s)
+        if isinstance(lit, (list, tuple)):
+            return list(lit)
+    except Exception:
+        pass
+    # fallback comma separated
+    s2 = s.strip("[]() ")
+    parts = [p.strip() for p in s2.split(",") if p.strip() != ""]
+    return parts
+
+
+def _parse_numeric(v: Any, fallback: Any = None) -> Any:
+    if v is None:
+        return fallback
+    if isinstance(v, (int, float)):
+        return v
+    s = str(v).strip()
+    if s == "" or s.lower() == "none":
+        return fallback
+    try:
+        return int(s)
+    except Exception:
+        try:
+            return float(s)
+        except Exception:
+            return fallback
+
+def _try_json_or_literal(s: Any) -> Any:
+    """Try parse JSON or python literal; otherwise return original string."""
+    if s is None:
+        return None
+    if not isinstance(s, str):
+        return s
+    s0 = s.strip()
+    if s0 == "":
+        return None
+    # try json
+    try:
+        return json.loads(s0)
+    except Exception:
+        pass
+    # try python literal
+    try:
+        return ast.literal_eval(s0)
+    except Exception:
+        pass
+    return s
+
+
+def resolve_aligner_args(
+    merged: dict,
+    default_by_aligner: Optional[Dict[str, List[str]]] = None,
+    aligner_synonyms: Optional[Dict[str, str]] = None,
+) -> List[str]:
+    """
+    Resolve merged['aligner_args'] into a concrete list for the chosen aligner and sequencer.
+
+    Behavior (search order):
+      1. If aligner_args is a dict, try keys in this order (case-insensitive):
+          a) "<aligner>@<sequencer>" (top-level combined key)
+          b) aligner -> (if dict) sequencer (nested) -> 'default' fallback
+          c) aligner -> (if list) use that list
+          d) top-level 'default' key in aligner_args dict
+      2. If aligner_args is a list -> return it (applies to any aligner/sequencer).
+      3. If aligner_args is a string -> try parse JSON/literal or return single-element list.
+      4. Otherwise fall back to builtin defaults per aligner.
+    """
+    # builtin defaults (aligner -> args)
+    builtin_defaults = {
+        "minimap2": ['-a', '-x', 'map-ont', '--MD', '-Y', '-y', '-N', '5', '--secondary=no'],
+        "dorado": ['--mm2-opts', '-N', '5'],
+    }
+    if default_by_aligner is None:
+        default_by_aligner = builtin_defaults
+
+    # synonyms mapping
+    synonyms = {"mm2": "minimap2", "minimap": "minimap2", "minimap-2": "minimap2"}
+    if aligner_synonyms:
+        synonyms.update(aligner_synonyms)
+
+    # canonicalize requested aligner and sequencer
+    raw_aligner = merged.get("aligner", "minimap2") or "minimap2"
+    raw_sequencer = merged.get("sequencer", None)  # e.g. 'ont', 'pacbio', 'illumina'
+    key_align = str(raw_aligner).strip().lower()
+    key_seq = None if raw_sequencer is None else str(raw_sequencer).strip().lower()
+    if key_align in synonyms:
+        key_align = synonyms[key_align]
+
+    raw = merged.get("aligner_args", None)
+
+    # helper to coerce a candidate to list[str]
+    def _coerce_to_list(val):
+        if isinstance(val, (list, tuple)):
+            return [str(x) for x in val]
+        if isinstance(val, str):
+            parsed = _try_json_or_literal(val)
+            if isinstance(parsed, (list, tuple)):
+                return [str(x) for x in parsed]
+            return [str(parsed)]
+        if val is None:
+            return None
+        return [str(val)]
+
+    # If dict, do layered lookups
+    if isinstance(raw, dict):
+        # case-insensitive dict
+        top_map = {str(k).lower(): v for k, v in raw.items()}
+
+        # 1) try combined top-level key "aligner@sequencer"
+        if key_seq:
+            combined_key = f"{key_align}@{key_seq}"
+            if combined_key in top_map:
+                res = _coerce_to_list(top_map[combined_key])
+                if res:
+                    return res
+
+        # 2) try aligner key
+        if key_align in top_map:
+            val = top_map[key_align]
+            # if nested dict: try sequencer key then 'default'
+            if isinstance(val, dict):
+                submap = {str(k).lower(): v for k, v in val.items()}
+                if key_seq and key_seq in submap:
+                    res = _coerce_to_list(submap[key_seq])
+                    if res:
+                        return res
+                if "default" in submap:
+                    res = _coerce_to_list(submap["default"])
+                    if res:
+                        return res
+                # nothing matched inside aligner->dict; fall back to top-level aligner (no sequencer)
+            else:
+                # aligner maps to list/str: use it
+                res = _coerce_to_list(val)
+                if res:
+                    return res
+
+        # 3) try top-level 'default' key inside aligner_args mapping
+        if "default" in top_map:
+            res = _coerce_to_list(top_map["default"])
+            if res:
+                return res
+
+        # 4) last top-level attempt: any key equal to aligner synonyms etc (already handled)
+        # fallthrough to builtin
+    # If user provided a concrete list -> use it
+    if isinstance(raw, (list, tuple)):
+        return [str(x) for x in raw]
+
+    # If scalar string, attempt to parse
+    if isinstance(raw, str):
+        parsed = _try_json_or_literal(raw)
+        if isinstance(parsed, (list, tuple)):
+            return [str(x) for x in parsed]
+        return [str(parsed)]
+
+    # Nothing found -> fallback builtin default
+    return list(default_by_aligner.get(key_align, []))
+
+
+# HMM default params and hepler functions
+def normalize_hmm_feature_sets(raw: Any) -> Dict[str, dict]:
+    """
+    Normalize user-provided `hmm_feature_sets` into canonical structure:
+      { group_name: {"features": {label: (lo, hi), ...}, "state": "<Modified|Non-Modified>"} }
+    Accepts dict, JSON/string, None. Returns {} for empty input.
+    """
+    if raw is None:
+        return {}
+    parsed = raw
+    if isinstance(raw, str):
+        parsed = _try_json_or_literal(raw)
+    if not isinstance(parsed, dict):
+        return {}
+
+    def _coerce_bound(x):
+        if x is None:
+            return None
+        if isinstance(x, (int, float)):
+            return float(x)
+        s = str(x).strip().lower()
+        if s in ("inf", "infty", "infinite"):
+            return np.inf
+        if s in ("none", ""):
+            return None
+        try:
+            return float(x)
+        except Exception:
+            return None
+
+    def _coerce_feature_map(feats):
+        out = {}
+        if not isinstance(feats, dict):
+            return out
+        for fname, rng in feats.items():
+            if rng is None:
+                out[fname] = (0.0, np.inf)
+                continue
+            if isinstance(rng, (list, tuple)) and len(rng) >= 2:
+                lo = _coerce_bound(rng[0]) or 0.0
+                hi = _coerce_bound(rng[1])
+                if hi is None:
+                    hi = np.inf
+                out[fname] = (float(lo), float(hi) if not np.isinf(hi) else np.inf)
+            else:
+                # scalar -> treat as upper bound
+                val = _coerce_bound(rng)
+                out[fname] = (0.0, float(val) if val is not None else np.inf)
+        return out
+
+    canonical = {}
+    for grp, info in parsed.items():
+        if not isinstance(info, dict):
+            feats = _coerce_feature_map(info)
+            canonical[grp] = {"features": feats, "state": "Modified"}
+            continue
+        feats = _coerce_feature_map(info.get("features", info.get("ranges", {})))
+        state = info.get("state", info.get("label", "Modified"))
+        canonical[grp] = {"features": feats, "state": state}
+    return canonical
+
+
+# -------------------------
+# LoadExperimentConfig
+# -------------------------
+class LoadExperimentConfig:
+    """
+    Load an experiment CSV (or DataFrame / file-like) into a typed var_dict.
+
+    CSV expected columns: 'variable', 'value', optional 'type'.
+    If 'type' missing, the loader will infer type.
+
+    Example
+    -------
+    loader = LoadExperimentConfig("experiment_config.csv")
+    var_dict = loader.var_dict
+    """
+
+    def __init__(self, experiment_config: Union[str, Path, IO, pd.DataFrame]):
+        self.source = experiment_config
+        self.df = self._load_df(experiment_config)
+        self.var_dict = self._parse_df(self.df)
+
+    @staticmethod
+    def _load_df(source: Union[str, Path, IO, pd.DataFrame]) -> pd.DataFrame:
+        """Load a pandas DataFrame from path, file-like, or accept if already DataFrame."""
+        if isinstance(source, pd.DataFrame):
+            df = source.copy()
+        else:
+            if isinstance(source, (str, Path)):
+                p = Path(source)
+                if not p.exists():
+                    raise FileNotFoundError(f"Config file not found: {source}")
+                df = pd.read_csv(p, dtype=str, keep_default_na=False, na_values=[""])
+            else:
+                # file-like
+                df = pd.read_csv(source, dtype=str, keep_default_na=False, na_values=[""])
+        # normalize column names
+        df.columns = [c.strip() for c in df.columns]
+        if 'variable' not in df.columns:
+            raise ValueError("Config CSV must contain a 'variable' column.")
+        if 'value' not in df.columns:
+            df['value'] = ''
+        if 'type' not in df.columns:
+            df['type'] = ''
+        return df
+
+    @staticmethod
+    def _parse_value_as_type(value_str: Optional[str], dtype_hint: Optional[str]) -> Any:
+        """
+        Parse a single value string into a Python object guided by dtype_hint (or infer).
+        Supports int, float, bool, list, JSON, Python literal, or string.
+        """
+        if value_str is None:
+            return None
+        v = str(value_str).strip()
+        if v == "" or v.lower() == "none":
+            return None
+
+        hint = "" if dtype_hint is None else str(dtype_hint).strip().lower()
+
+        def parse_bool(s: str):
+            s2 = s.strip().lower()
+            if s2 in ('1', 'true', 't', 'yes', 'y', 'on'):
+                return True
+            if s2 in ('0', 'false', 'f', 'no', 'n', 'off'):
+                return False
+            raise ValueError(f"Cannot parse boolean from '{s}'")
+
+        def parse_list_like(s: str):
+            # try JSON first
+            try:
+                val = json.loads(s)
+                if isinstance(val, list):
+                    return val
+            except Exception:
+                pass
+            # try python literal
+            try:
+                val = ast.literal_eval(s)
+                if isinstance(val, (list, tuple)):
+                    return list(val)
+            except Exception:
+                pass
+            # fallback split
+            parts = [p.strip() for p in s.strip("()[] ").split(',') if p.strip() != ""]
+            return parts
+
+        if hint in ('int', 'integer'):
+            return int(v)
+        if hint in ('float', 'double'):
+            return float(v)
+        if hint in ('bool', 'boolean'):
+            return parse_bool(v)
+        if hint in ('list', 'array'):
+            return parse_list_like(v)
+        if hint in ('string', 'str'):
+            return v
+
+        # infer
+        try:
+            return int(v)
+        except Exception:
+            pass
+        try:
+            return float(v)
+        except Exception:
+            pass
+        try:
+            return parse_bool(v)
+        except Exception:
+            pass
+        try:
+            j = json.loads(v)
+            return j
+        except Exception:
+            pass
+        try:
+            lit = ast.literal_eval(v)
+            return lit
+        except Exception:
+            pass
+        if (',' in v) and (not any(ch in v for ch in '{}[]()')):
+            return [p.strip() for p in v.split(',') if p.strip() != ""]
+        return v
+
+    def _parse_df(self, df: pd.DataFrame) -> Dict[str, Any]:
+        parsed: Dict[str, Any] = {}
+        for idx, row in df.iterrows():
+            name = str(row['variable']).strip()
+            if name == "":
+                continue
+            raw_val = row.get('value', "")
+            raw_type = row.get('type', "")
+            if pd.isna(raw_val) or str(raw_val).strip() == "":
+                raw_val = None
+            try:
+                parsed_val = self._parse_value_as_type(raw_val, raw_type)
+            except Exception as e:
+                warnings.warn(f"Failed to parse config variable '{name}' (row {idx}): {e}. Storing raw value.")
+                parsed_val = None if raw_val is None else raw_val
+            if name in parsed:
+                warnings.warn(f"Duplicate config variable '{name}' encountered (row {idx}). Overwriting previous value.")
+            parsed[name] = parsed_val
+        return parsed
+
+    def to_dataframe(self) -> pd.DataFrame:
+        """Return parsed config as a pandas DataFrame (variable, value)."""
+        rows = []
+        for k, v in self.var_dict.items():
+            rows.append({'variable': k, 'value': v})
+        return pd.DataFrame(rows)
+
+
+# -------------------------
+# deep merge & defaults loader (with inheritance)
+# -------------------------
+def deep_merge(a: Dict[str, Any], b: Dict[str, Any]) -> Dict[str, Any]:
+    """
+    Recursively merge two dicts: returns new dict = a merged with b, where b overrides.
+    If both values are dicts -> merge recursively; else b replaces a.
+    """
+    out = dict(a or {})
+    for k, v in (b or {}).items():
+        if k in out and isinstance(out[k], dict) and isinstance(v, dict):
+            out[k] = deep_merge(out[k], v)
+        else:
+            out[k] = v
+    return out
+
+
+def _load_defaults_file(path: Path) -> Dict[str, Any]:
+    if not path.exists():
+        return {}
+    text = path.read_text(encoding="utf8")
+    suffix = path.suffix.lower()
+    if suffix in (".yaml", ".yml"):
+        if yaml is None:
+            raise RuntimeError("PyYAML required to load YAML defaults (pip install pyyaml).")
+        return yaml.safe_load(text) or {}
+    elif suffix == ".json":
+        return json.loads(text or "{}")
+    else:
+        # try json then yaml if available
+        try:
+            return json.loads(text)
+        except Exception:
+            if yaml is not None:
+                return yaml.safe_load(text) or {}
+            raise RuntimeError(f"Unknown defaults file type for {path}; provide JSON or YAML.")
+
+
+def load_defaults_with_inheritance(
+    defaults_dir: Union[str, Path],
+    modality: Optional[str],
+    *,
+    default_basename: str = "default",
+    allowed_exts: Tuple[str, ...] = (".yaml", ".yml", ".json"),
+    debug: bool = False,
+) -> Tuple[Dict[str, Any], List[str]]:
+    """
+    Strict loader: only loads default + modality + any explicit 'extends' chain.
+
+    - defaults_dir: directory containing defaults files.
+    - modality: name of modality (e.g. "GpC"). We look for <modality>.<ext> in defaults_dir.
+    - default_basename: name of fallback default file (without extension).
+    - allowed_exts: allowed extensions to try.
+    - debug: if True, prints what was loaded.
+
+    Returns (merged_defaults_dict, load_order_list) where load_order_list are resolved file paths read.
+    """
+    pdir = Path(defaults_dir) if defaults_dir is not None else None
+    if pdir is None or not pdir.exists():
+        return {}, []
+
+    # Resolve a "name" to a file in defaults_dir.
+    # Only treat `name` as an explicit path if it contains a path separator or is absolute.
+    def resolve_name_to_path(name: str) -> Optional[Path]:
+        n = str(name).strip()
+        if n == "":
+            return None
+        cand = Path(n)
+        # If user provided a path-like string (contains slash/backslash or absolute), allow it
+        if cand.is_absolute() or ("/" in n) or ("\\" in n):
+            if cand.exists() and cand.suffix.lower() in allowed_exts:
+                return cand.resolve()
+            return None
+        # Otherwise only look inside defaults_dir for name + ext (do NOT treat bare name as arbitrary file)
+        for ext in allowed_exts:
+            p = pdir / f"{n}{ext}"
+            if p.exists():
+                return p.resolve()
+        return None
+
+    visited = set()
+    load_order: List[str] = []
+
+    def _rec_load(name_or_path: Union[str, Path]) -> Dict[str, Any]:
+        # Resolve to a file path (strict)
+        if isinstance(name_or_path, Path):
+            p = name_or_path
+        else:
+            p = resolve_name_to_path(str(name_or_path))
+        if p is None:
+            if debug:
+                print(f"[defaults loader] resolve failed for '{name_or_path}'")
+            return {}
+        p = Path(p).resolve()
+        p_str = str(p)
+        if p_str in visited:
+            if debug:
+                print(f"[defaults loader] already visited {p_str} (skipping to avoid cycle)")
+            return {}
+        visited.add(p_str)
+
+        data = _load_defaults_file(p)  # reuse your existing helper
+        if not isinstance(data, dict):
+            if debug:
+                print(f"[defaults loader] file {p_str} did not produce a dict -> ignoring")
+            data = {}
+
+        # Extract any extends/inherits keys (string or list). They reference other named default files.
+        bases = []
+        for key in ("extends", "inherits", "base"):
+            if key in data:
+                b = data.pop(key)
+                if isinstance(b, (list, tuple)):
+                    bases = list(b)
+                elif isinstance(b, str):
+                    bases = [b]
+                break
+
+        merged = {}
+        # Load bases first (in order); bases are resolved relative to defaults_dir unless given as path
+        for base_name in bases:
+            base_defaults = _rec_load(base_name)
+            merged = deep_merge(merged, base_defaults)
+
+        # Then merge this file's data (this file overrides its bases)
+        merged = deep_merge(merged, data)
+        load_order.append(p_str)
+        if debug:
+            print(f"[defaults loader] loaded {p_str}")
+        return merged
+
+    merged_defaults = {}
+    # Load default.* first if present
+    def_path = resolve_name_to_path(default_basename)
+    if def_path is not None:
+        merged_defaults = deep_merge(merged_defaults, _rec_load(def_path))
+
+    # Load modality.* if present (modality overrides default)
+    if modality:
+        mod_path = resolve_name_to_path(modality)
+        if mod_path is not None:
+            merged_defaults = deep_merge(merged_defaults, _rec_load(mod_path))
+        else:
+            if debug:
+                print(f"[defaults loader] no modality file found for '{modality}' in {pdir}")
+
+    if debug:
+        print("[defaults loader] final load order:", load_order)
+    return merged_defaults, load_order
+
+
+# -------------------------
+# ExperimentConfig dataclass
+# -------------------------
+@dataclass
+class ExperimentConfig:
+    # Compute
+    threads: Optional[int] = None
+    device: str = "auto"
+
+    # General I/O
+    input_data_path: Optional[str] = None
+    output_directory: Optional[str] = None
+    fasta: Optional[str] = None
+    bam_suffix: str = ".bam"
+    recursive_input_search: bool = True
+    split_dir: str = "demultiplexed_BAMs"
+    strands: List[str] = field(default_factory=lambda: ["bottom", "top"])
+    conversions: List[str] = field(default_factory=lambda: ["unconverted"])
+    fasta_regions_of_interest: Optional[str] = None
+    sample_sheet_path: Optional[str] = None
+    sample_sheet_mapping_column: Optional[str] = 'Barcode'
+    experiment_name: Optional[str] = None
+    input_already_demuxed: bool = False
+
+    # FASTQ input specific
+    fastq_barcode_map: Optional[Dict[str, str]] = None
+    fastq_auto_pairing: bool = True
+
+    # Conversion/Deamination file handling
+    delete_intermediate_hdfs: bool = True
+
+    # Direct SMF specific params for initial AnnData loading
+    batch_size: int = 4
+    skip_unclassified: bool = True
+    delete_batch_hdfs: bool = True
+
+    # Sequencing modality and general experiment params
+    smf_modality: Optional[str] = None
+    sequencer: Optional[str] = None
+
+    # Enzyme / mod targets
+    mod_target_bases: List[str] = field(default_factory=lambda: ["GpC", "CpG"])
+    enzyme_target_bases: List[str] = field(default_factory=lambda: ["GpC"])
+
+    # Conversion/deamination
+    conversion_types: List[str] = field(default_factory=lambda: ["5mC"])
+
+    # Nanopore specific for basecalling and demultiplexing
+    model_dir: Optional[str] = None
+    barcode_kit: Optional[str] = None
+    model: str = "hac"
+    barcode_both_ends: bool = False
+    trim: bool = False
+    # General basecalling params
+    filter_threshold: float = 0.8
+    # Modified basecalling specific params
+    m6A_threshold: float = 0.7
+    m5C_threshold: float = 0.7
+    hm5C_threshold: float = 0.7
+    thresholds: List[float] = field(default_factory=list)
+    mod_list: List[str] = field(default_factory=lambda: ["5mC_5hmC", "6mA"])
+
+    # Alignment params
+    mapping_threshold: float = 0.01 # Min threshold for fraction of reads in a sample mapping to a reference in order to include the reference in the anndata
+    aligner: str = "minimap2"
+    aligner_args: Optional[List[str]] = None
+    make_bigwigs: bool = False
+
+    # Anndata structure
+    reference_column: Optional[str] = 'Reference_strand'
+    sample_column: Optional[str] = 'Barcode'
+
+    # General Plotting
+    sample_name_col_for_plotting: Optional[str] = 'Barcode'
+    rows_per_qc_histogram_grid: int = 12
+
+    # Preprocessing - Read length and quality filter params
+    read_coord_filter: Optional[Sequence[float]] = field(default_factory=lambda: [None, None])
+    read_len_filter_thresholds: Optional[Sequence[float]] = field(default_factory=lambda: [200, None])
+    read_len_to_ref_ratio_filter_thresholds: Optional[Sequence[float]] = field(default_factory=lambda: [0.4, 1.1])
+    read_quality_filter_thresholds: Optional[Sequence[float]] = field(default_factory=lambda: [20, None])
+    read_mapping_quality_filter_thresholds: Optional[Sequence[float]] = field(default_factory=lambda: [None, None])
+
+    # Preprocessing - Read modification filter params
+    read_mod_filtering_gpc_thresholds: List[float] = field(default_factory=lambda: [0.025, 0.975])
+    read_mod_filtering_cpg_thresholds: List[float] = field(default_factory=lambda: [0.00, 1])
+    read_mod_filtering_any_c_thresholds: List[float] = field(default_factory=lambda: [0.025, 0.975])
+    read_mod_filtering_a_thresholds: List[float] = field(default_factory=lambda: [0.025, 0.975])
+    read_mod_filtering_use_other_c_as_background: bool = True
+    min_valid_fraction_positions_in_read_vs_ref: float = 0.2
+
+    # Preprocessing - Duplicate detection params
+    duplicate_detection_site_types: List[str] = field(default_factory=lambda: ['GpC', 'CpG', 'ambiguous_GpC_CpG'])
+    duplicate_detection_distance_threshold: float = 0.07
+    hamming_vs_metric_keys: List[str] = field(default_factory=lambda: ['Fraction_any_C_site_modified'])
+    duplicate_detection_keep_best_metric: str ='read_quality'
+    duplicate_detection_window_size_for_hamming_neighbors: int = 50
+    duplicate_detection_min_overlapping_positions: int = 20
+    duplicate_detection_do_hierarchical: bool = True
+    duplicate_detection_hierarchical_linkage: str = "average"
+    duplicate_detection_do_pca: bool = False
+
+    # Preprocessing - Complexity analysis params
+
+    # Basic Analysis - Clustermap params
+    layer_for_clustermap_plotting: Optional[str] = 'nan0_0minus1'
+
+    # Basic Analysis - UMAP/Leiden params
+    layer_for_umap_plotting: Optional[str] = 'nan_half'
+    umap_layers_to_plot: List[str] = field(default_factory=lambda: ["mapped_length", "Raw_modification_signal"]) 
+
+    # Basic Analysis - Spatial Autocorrelation params
+    rows_per_qc_autocorr_grid: int = 12
+    autocorr_rolling_window_size: int = 25
+    autocorr_max_lag: int = 800
+    autocorr_site_types: List[str] = field(default_factory=lambda: ['GpC', 'CpG', 'any_C'])
+
+    # Basic Analysis - Correlation Matrix params
+    correlation_matrix_types: List[str] = field(default_factory=lambda: ["pearson", "binary_covariance"]) 
+    correlation_matrix_cmaps: List[str] = field(default_factory=lambda: ["seismic", "viridis"]) 
+    correlation_matrix_site_types: List[str] = field(default_factory=lambda: ["GpC_site"]) 
+
+    # HMM params
+    hmm_n_states: int = 2
+    hmm_init_emission_probs: List[list] = field(default_factory=lambda: [[0.8, 0.2], [0.2, 0.8]]) 
+    hmm_init_transition_probs: List[list] = field(default_factory=lambda: [[0.9, 0.1], [0.1, 0.9]]) 
+    hmm_init_start_probs: List[float] = field(default_factory=lambda: [0.5, 0.5])
+    hmm_eps: float = 1e-8
+    hmm_dtype: str = "float64"
+    hmm_annotation_threshold: float = 0.5
+    hmm_batch_size: int = 1024
+    hmm_use_viterbi: bool = False
+    hmm_device: Optional[str] = None
+    hmm_methbases: Optional[List[str]] = None  # if None, HMM.annotate_adata will fall back to mod_target_bases
+    footprints: Optional[bool] = True
+    accessible_patches: Optional[bool] = True
+    cpg: Optional[bool] = False
+    hmm_feature_sets: Dict[str, Any] = field(default_factory=dict)
+    hmm_merge_layer_features: Optional[List[Tuple]] = field(default_factory=lambda: [(None, 80)])
+
+    # Pipeline control flow - preprocessing and QC
+    force_redo_preprocessing: bool = False
+    force_reload_sample_sheet: bool = True
+    bypass_add_read_length_and_mapping_qc: bool = False
+    force_redo_add_read_length_and_mapping_qc: bool = False
+    bypass_clean_nan: bool = False
+    force_redo_clean_nan: bool = False
+    bypass_append_base_context: bool = False
+    force_redo_append_base_context: bool = False
+    invert_adata: bool = False
+    bypass_append_binary_layer_by_base_context: bool = False
+    force_redo_append_binary_layer_by_base_context: bool = False
+    bypass_calculate_read_modification_stats: bool = False
+    force_redo_calculate_read_modification_stats: bool = False
+    bypass_filter_reads_on_modification_thresholds: bool = False
+    force_redo_filter_reads_on_modification_thresholds: bool = False
+    bypass_flag_duplicate_reads: bool = False
+    force_redo_flag_duplicate_reads: bool = False
+    bypass_complexity_analysis: bool = False 
+    force_redo_complexity_analysis: bool = False
+
+    # Pipeline control flow - Basic Analyses
+    force_redo_basic_analyses: bool = False
+    bypass_basic_clustermaps: bool = False
+    force_redo_basic_clustermaps: bool = False
+    bypass_basic_umap: bool = False
+    force_redo_basic_umap: bool = False
+    bypass_spatial_autocorr_calculations: bool = False
+    force_redo_spatial_autocorr_calculations: bool = False
+    bypass_spatial_autocorr_plotting: bool = False
+    force_redo_spatial_autocorr_plotting: bool = False
+    bypass_matrix_corr_calculations: bool = False
+    force_redo_matrix_corr_calculations: bool = False
+    bypass_matrix_corr_plotting: bool = False
+    force_redo_matrix_corr_plotting: bool = False
+
+    # Pipeline control flow - HMM Analyses
+    bypass_hmm_fit: bool = False
+    force_redo_hmm_fit: bool = False
+    bypass_hmm_apply: bool = False
+    force_redo_hmm_apply: bool = False
+
+    # metadata
+    config_source: Optional[str] = None
+
+    # -------------------------
+    # Construction helpers
+    # -------------------------
+    @classmethod
+    def from_var_dict(
+        cls,
+        var_dict: Optional[Dict[str, Any]],
+        date_str: Optional[str] = None,
+        config_source: Optional[str] = None,
+        defaults_dir: Optional[Union[str, Path]] = None,
+        defaults_map: Optional[Dict[str, Dict[str, Any]]] = None,
+        merge_with_defaults: bool = True,
+        override_with_csv: bool = True,
+        allow_csv_extends: bool = True,
+        allow_null_override: bool = False,
+    ) -> Tuple["ExperimentConfig", Dict[str, Any]]:
+        """
+        Create ExperimentConfig from a raw var_dict (as produced by LoadExperimentConfig).
+        Returns (instance, report) where report contains modality/defaults/merged info.
+
+        merge_with_defaults: load defaults from defaults_dir or defaults_map.
+        override_with_csv: CSV values override defaults; if False defaults take precedence.
+        allow_csv_extends: allow the CSV to include 'extends' to pull in extra defaults files.
+        allow_null_override: if False, CSV keys with value None will NOT override defaults (keeps defaults).
+        """
+        var_dict = var_dict or {}
+
+        # 1) normalize incoming values
+        normalized: Dict[str, Any] = {}
+        for k, v in var_dict.items():
+            if v is None:
+                normalized[k] = None
+                continue
+            if isinstance(v, str):
+                s = v.strip()
+                if s == "" or s.lower() == "none":
+                    normalized[k] = None
+                else:
+                    normalized[k] = _try_json_or_literal(s)
+            else:
+                normalized[k] = v
+
+        modality = normalized.get("smf_modality")
+        if isinstance(modality, (list, tuple)) and len(modality) > 0:
+            modality = modality[0]
+
+        defaults_loaded = {}
+        defaults_source_chain: List[str] = []
+        if merge_with_defaults:
+            if defaults_map and modality in defaults_map:
+                defaults_loaded = dict(defaults_map[modality] or {})
+                defaults_source_chain = [f"defaults_map['{modality}']"]
+            elif defaults_dir is not None:
+                defaults_loaded, defaults_source_chain = load_defaults_with_inheritance(defaults_dir, modality)
+
+        # If CSV asks to extend defaults, load those and merge
+        merged = dict(defaults_loaded or {})
+
+        if allow_csv_extends:
+            extends = normalized.get("extends") or normalized.get("inherits")
+            if extends:
+                if isinstance(extends, str):
+                    ext_list = [extends]
+                elif isinstance(extends, (list, tuple)):
+                    ext_list = list(extends)
+                else:
+                    ext_list = []
+                for ext in ext_list:
+                    ext_defaults, ext_sources = (load_defaults_with_inheritance(defaults_dir, ext) if defaults_dir else ({}, []))
+                    merged = deep_merge(merged, ext_defaults)
+                    for s in ext_sources:
+                        if s not in defaults_source_chain:
+                            defaults_source_chain.append(s)
+
+        # Now overlay CSV values
+        # Prepare csv_effective depending on allow_null_override
+        csv_effective = {}
+        for k, v in normalized.items():
+            if k in ("extends", "inherits"):
+                continue
+            if v is None and not allow_null_override:
+                # skip: keep default
+                continue
+            csv_effective[k] = v
+
+        if override_with_csv:
+            merged = deep_merge(merged, csv_effective)
+        else:
+            # defaults take precedence: only set keys missing in merged
+            for k, v in csv_effective.items():
+                if k not in merged:
+                    merged[k] = v
+
+        # experiment_name default
+        if merged.get("experiment_name") is None and date_str:
+            merged["experiment_name"] = f"{date_str}_SMF_experiment"
+
+        # final normalization
+        if "strands" in merged:
+            merged["strands"] = _parse_list(merged["strands"])
+        if "conversions" in merged:
+            merged["conversions"] = _parse_list(merged["conversions"])
+        if "mod_target_bases" in merged:
+            merged["mod_target_bases"] = _parse_list(merged["mod_target_bases"])
+        if "conversion_types" in merged:
+            merged["conversion_types"] = _parse_list(merged["conversion_types"])
+
+        merged["filter_threshold"] = float(_parse_numeric(merged.get("filter_threshold", 0.8), 0.8))
+        merged["m6A_threshold"] = float(_parse_numeric(merged.get("m6A_threshold", 0.7), 0.7))
+        merged["m5C_threshold"] = float(_parse_numeric(merged.get("m5C_threshold", 0.7), 0.7))
+        merged["hm5C_threshold"] = float(_parse_numeric(merged.get("hm5C_threshold", 0.7), 0.7))
+        merged["thresholds"] = [
+            merged["filter_threshold"],
+            merged["m6A_threshold"],
+            merged["m5C_threshold"],
+            merged["hm5C_threshold"],
+        ]
+
+        for bkey in ("barcode_both_ends", "trim", "input_already_demuxed", "make_bigwigs", "skip_unclassified", "delete_batch_hdfs"):
+            if bkey in merged:
+                merged[bkey] = _parse_bool(merged[bkey])
+
+        if "batch_size" in merged:
+            merged["batch_size"] = int(_parse_numeric(merged.get("batch_size", 4), 4))
+        if "threads" in merged:
+            tval = _parse_numeric(merged.get("threads", None), None)
+            merged["threads"] = None if tval is None else int(tval)
+                
+        if "aligner_args" in merged and merged.get("aligner_args") is None:
+            merged.pop("aligner_args", None)
+
+        # --- Resolve aligner_args into concrete list for the chosen aligner ---
+        merged['aligner_args'] = resolve_aligner_args(merged)
+
+        if "mod_list" in merged:
+            merged["mod_list"] = _parse_list(merged.get("mod_list"))
+
+        # HMM feature set handling
+        if "hmm_feature_sets" in merged:
+            merged["hmm_feature_sets"] = normalize_hmm_feature_sets(merged["hmm_feature_sets"])
+        else:
+            # allow older names (footprint_ranges, accessible_ranges, cpg_ranges) — optional:
+            maybe_fs = {}
+            if "footprint_ranges" in merged or "hmm_footprint_ranges" in merged:
+                maybe_fs["footprint"] = {"features": merged.get("hmm_footprint_ranges", merged.get("footprint_ranges")), "state": merged.get("hmm_footprint_state", "Non-Modified")}
+            if "accessible_ranges" in merged or "hmm_accessible_ranges" in merged:
+                maybe_fs["accessible"] = {"features": merged.get("hmm_accessible_ranges", merged.get("accessible_ranges")), "state": merged.get("hmm_accessible_state", "Modified")}
+            if "cpg_ranges" in merged or "hmm_cpg_ranges" in merged:
+                maybe_fs["cpg"] = {"features": merged.get("hmm_cpg_ranges", merged.get("cpg_ranges")), "state": merged.get("hmm_cpg_state", "Modified")}
+            if maybe_fs:
+                merged.setdefault("hmm_feature_sets", {})
+                for k, v in maybe_fs.items():
+                    merged["hmm_feature_sets"].setdefault(k, v)
+
+            # final normalization will be done below
+            # (do not set local hmm_feature_sets here — do it once below)
+            pass
+
+        # Final normalization of hmm_feature_sets and canonical local variables
+        merged["hmm_feature_sets"] = normalize_hmm_feature_sets(merged.get("hmm_feature_sets", {}))
+        hmm_feature_sets = merged.get("hmm_feature_sets", {})
+        hmm_annotation_threshold = merged.get("hmm_annotation_threshold", 0.5)
+        hmm_batch_size = int(merged.get("hmm_batch_size", 1024))
+        hmm_use_viterbi = bool(merged.get("hmm_use_viterbi", False))
+        hmm_device = merged.get("hmm_device", None)
+        hmm_methbases = _parse_list(merged.get("hmm_methbases", None))
+        if not hmm_methbases:  # None or []
+            hmm_methbases = _parse_list(merged.get("mod_target_bases", None))
+        if not hmm_methbases:
+            hmm_methbases = ['C']
+        hmm_methbases = list(hmm_methbases)
+        hmm_merge_layer_features = _parse_list(merged.get("hmm_merge_layer_features", None))
+
+
+        # instantiate dataclass
+        instance = cls(
+            smf_modality = merged.get("smf_modality"),
+            input_data_path = merged.get("input_data_path"),
+            recursive_input_search = merged.get("recursive_input_search"),
+            output_directory = merged.get("output_directory"),
+            fasta = merged.get("fasta"),
+            sequencer = merged.get("sequencer"),
+            model_dir = merged.get("model_dir"),
+            barcode_kit = merged.get("barcode_kit"),
+            fastq_barcode_map = merged.get("fastq_barcode_map"),
+            fastq_auto_pairing = merged.get("fastq_auto_pairing"),
+            bam_suffix = merged.get("bam_suffix", ".bam"),
+            split_dir = merged.get("split_dir", "demultiplexed_BAMs"),
+            strands = merged.get("strands", ["bottom","top"]),
+            conversions = merged.get("conversions", ["unconverted"]),
+            fasta_regions_of_interest = merged.get("fasta_regions_of_interest"),
+            mapping_threshold = float(merged.get("mapping_threshold", 0.01)),
+            experiment_name = merged.get("experiment_name"),
+            model = merged.get("model", "hac"),
+            barcode_both_ends = merged.get("barcode_both_ends", False),
+            trim = merged.get("trim", False),
+            input_already_demuxed = merged.get("input_already_demuxed", False),
+            threads = merged.get("threads"),
+            sample_sheet_path = merged.get("sample_sheet_path"),
+            sample_sheet_mapping_column = merged.get("sample_sheet_mapping_column"),
+            aligner = merged.get("aligner", "minimap2"),
+            aligner_args = merged.get("aligner_args", None),
+            device = merged.get("device", "auto"),
+            make_bigwigs = merged.get("make_bigwigs", False),
+            delete_intermediate_hdfs = merged.get("delete_intermediate_hdfs", True),
+            mod_target_bases = merged.get("mod_target_bases", ["GpC","CpG"]),
+            enzyme_target_bases = merged.get("enzyme_target_bases", ["GpC"]), 
+            conversion_types = merged.get("conversion_types", ["5mC"]),
+            filter_threshold = merged.get("filter_threshold", 0.8),
+            m6A_threshold = merged.get("m6A_threshold", 0.7),
+            m5C_threshold = merged.get("m5C_threshold", 0.7),
+            hm5C_threshold = merged.get("hm5C_threshold", 0.7),
+            thresholds = merged.get("thresholds", []),
+            mod_list = merged.get("mod_list", ["5mC_5hmC","6mA"]),
+            batch_size = merged.get("batch_size", 4),
+            skip_unclassified = merged.get("skip_unclassified", True),
+            delete_batch_hdfs = merged.get("delete_batch_hdfs", True),
+            reference_column = merged.get("reference_column", 'Reference_strand'),
+            sample_column = merged.get("sample_column", 'Barcode'),
+            sample_name_col_for_plotting = merged.get("sample_name_col_for_plotting", 'Barcode'),
+            layer_for_clustermap_plotting = merged.get("layer_for_clustermap_plotting", 'nan0_0minus1'), 
+            layer_for_umap_plotting = merged.get("layer_for_umap_plotting", 'nan_half'),
+            umap_layers_to_plot = merged.get("umap_layers_to_plot",["mapped_length", 'Raw_modification_signal']),
+            rows_per_qc_histogram_grid = merged.get("rows_per_qc_histogram_grid", 12),
+            rows_per_qc_autocorr_grid = merged.get("rows_per_qc_autocorr_grid", 12),
+            autocorr_rolling_window_size = merged.get("autocorr_rolling_window_size", 25),
+            autocorr_max_lag = merged.get("autocorr_max_lag", 800), 
+            autocorr_site_types = merged.get("autocorr_site_types", ['GpC', 'CpG', 'any_C']),
+            hmm_n_states = merged.get("hmm_n_states", 2), 
+            hmm_init_emission_probs = merged.get("hmm_init_emission_probs",[[0.8, 0.2], [0.2, 0.8]]),
+            hmm_init_transition_probs = merged.get("hmm_init_transition_probs",[[0.9, 0.1], [0.1, 0.9]]),
+            hmm_init_start_probs = merged.get("hmm_init_start_probs",[0.5, 0.5]),
+            hmm_eps = merged.get("hmm_eps", 1e-8),
+            hmm_dtype = merged.get("hmm_dtype", "float64"),
+            hmm_feature_sets = hmm_feature_sets,
+            hmm_annotation_threshold = hmm_annotation_threshold,
+            hmm_batch_size = hmm_batch_size,
+            hmm_use_viterbi = hmm_use_viterbi,
+            hmm_methbases = hmm_methbases,
+            hmm_device = hmm_device,
+            hmm_merge_layer_features = hmm_merge_layer_features,
+            footprints = merged.get("footprints", None),
+            accessible_patches = merged.get("accessible_patches", None),
+            cpg = merged.get("cpg", None),
+            read_coord_filter = merged.get("read_coord_filter", [None, None]), 
+            read_len_filter_thresholds = merged.get("read_len_filter_thresholds", [200, None]),
+            read_len_to_ref_ratio_filter_thresholds = merged.get("read_len_to_ref_ratio_filter_thresholds", [0.4, 1.1]),
+            read_quality_filter_thresholds = merged.get("read_quality_filter_thresholds", [20, None]),
+            read_mapping_quality_filter_thresholds = merged.get("read_mapping_quality_filter_thresholds", [None, None]),
+            read_mod_filtering_gpc_thresholds = merged.get("read_mod_filtering_gpc_thresholds", [0.025, 0.975]),
+            read_mod_filtering_cpg_thresholds = merged.get("read_mod_filtering_cpg_thresholds", [0.0, 1.0]),
+            read_mod_filtering_any_c_thresholds = merged.get("read_mod_filtering_any_c_thresholds", [0.025, 0.975]),
+            read_mod_filtering_a_thresholds = merged.get("read_mod_filtering_a_thresholds", [0.025, 0.975]),
+            read_mod_filtering_use_other_c_as_background = merged.get("read_mod_filtering_use_other_c_as_background", True),
+            min_valid_fraction_positions_in_read_vs_ref = merged.get("min_valid_fraction_positions_in_read_vs_ref", 0.2), 
+            duplicate_detection_site_types = merged.get("duplicate_detection_site_types", ['GpC', 'CpG', 'ambiguous_GpC_CpG']),
+            duplicate_detection_distance_threshold = merged.get("duplicate_detection_distance_threshold", 0.07),
+            duplicate_detection_keep_best_metric = merged.get("duplicate_detection_keep_best_metric", "read_quality"),
+            duplicate_detection_window_size_for_hamming_neighbors = merged.get("duplicate_detection_window_size_for_hamming_neighbors", 50),
+            duplicate_detection_min_overlapping_positions = merged.get("duplicate_detection_min_overlapping_positions", 20),
+            duplicate_detection_do_hierarchical = merged.get("duplicate_detection_do_hierarchical", True),
+            duplicate_detection_hierarchical_linkage = merged.get("duplicate_detection_hierarchical_linkage", "average"),
+            duplicate_detection_do_pca = merged.get("duplicate_detection_do_pca", False),
+            correlation_matrix_types = merged.get("correlation_matrix_types", ["pearson", "binary_covariance"]),
+            correlation_matrix_cmaps = merged.get("correlation_matrix_cmaps", ["seismic", "viridis"]),
+            correlation_matrix_site_types = merged.get("correlation_matrix_site_types", ["GpC_site"]),
+            hamming_vs_metric_keys = merged.get("hamming_vs_metric_keys", ['Fraction_any_C_site_modified']),
+            force_redo_preprocessing = merged.get("force_redo_preprocessing", False), 
+            force_reload_sample_sheet = merged.get("force_reload_sample_sheet", True),
+            bypass_add_read_length_and_mapping_qc = merged.get("bypass_add_read_length_and_mapping_qc", False),
+            force_redo_add_read_length_and_mapping_qc = merged.get("force_redo_add_read_length_and_mapping_qc", False),
+            bypass_clean_nan = merged.get("bypass_clean_nan", False),
+            force_redo_clean_nan = merged.get("force_redo_clean_nan", False),
+            bypass_append_base_context = merged.get("bypass_append_base_context", False),
+            force_redo_append_base_context = merged.get("force_redo_append_base_context", False),
+            invert_adata = merged.get("invert_adata", False),
+            bypass_append_binary_layer_by_base_context = merged.get("bypass_append_binary_layer_by_base_context", False),
+            force_redo_append_binary_layer_by_base_context = merged.get("force_redo_append_binary_layer_by_base_context", False),
+            bypass_calculate_read_modification_stats = merged.get("bypass_calculate_read_modification_stats", False),
+            force_redo_calculate_read_modification_stats = merged.get("force_redo_calculate_read_modification_stats", False),
+            bypass_filter_reads_on_modification_thresholds = merged.get("bypass_filter_reads_on_modification_thresholds", False),
+            force_redo_filter_reads_on_modification_thresholds = merged.get("force_redo_filter_reads_on_modification_thresholds", False),
+            bypass_flag_duplicate_reads = merged.get("bypass_flag_duplicate_reads", False),
+            force_redo_flag_duplicate_reads = merged.get("force_redo_flag_duplicate_reads", False),
+            bypass_complexity_analysis = merged.get("bypass_complexity_analysis", False),
+            force_redo_complexity_analysis = merged.get("force_redo_complexity_analysis", False),
+            force_redo_basic_analyses = merged.get("force_redo_basic_analyses", False),
+            bypass_basic_clustermaps = merged.get("bypass_basic_clustermaps", False),
+            force_redo_basic_clustermaps = merged.get("force_redo_basic_clustermaps", False),
+            bypass_basic_umap = merged.get("bypass_basic_umap", False),
+            force_redo_basic_umap = merged.get("force_redo_basic_umap", False),
+            bypass_spatial_autocorr_calculations = merged.get("bypass_spatial_autocorr_calculations", False),
+            force_redo_spatial_autocorr_calculations = merged.get("force_redo_spatial_autocorr_calculations", False),
+            bypass_spatial_autocorr_plotting = merged.get("bypass_spatial_autocorr_plotting", False),
+            force_redo_spatial_autocorr_plotting = merged.get("force_redo_spatial_autocorr_plotting", False),
+            bypass_matrix_corr_calculations = merged.get("bypass_matrix_corr_calculations", False),
+            force_redo_matrix_corr_calculations = merged.get("force_redo_matrix_corr_calculations", False),
+            bypass_matrix_corr_plotting = merged.get("bypass_matrix_corr_plotting", False),
+            force_redo_matrix_corr_plotting = merged.get("force_redo_matrix_corr_plotting", False),
+            bypass_hmm_fit = merged.get("bypass_hmm_fit", False),
+            force_redo_hmm_fit = merged.get("force_redo_hmm_fit", False),
+            bypass_hmm_apply = merged.get("bypass_hmm_apply", False),
+            force_redo_hmm_apply = merged.get("force_redo_hmm_apply", False),
+    
+            config_source = config_source or "<var_dict>",
+        )
+
+        report = {
+            "modality": modality,
+            "defaults_source_chain": defaults_source_chain,
+            "defaults_loaded": defaults_loaded,
+            "csv_normalized": normalized,
+            "merged": merged,
+        }
+        return instance, report
+
+    # convenience: load from CSV via LoadExperimentConfig
+    @classmethod
+    def from_csv(
+        cls,
+        csv_input: Union[str, Path, IO, pd.DataFrame],
+        date_str: Optional[str] = None,
+        config_source: Optional[str] = None,
+        defaults_dir: Optional[Union[str, Path]] = None,
+        defaults_map: Optional[Dict[str, Dict[str, Any]]] = None,
+        **kwargs,
+    ) -> Tuple["ExperimentConfig", Dict[str, Any]]:
+        """
+        Load CSV using LoadExperimentConfig (or accept DataFrame) and build ExperimentConfig.
+        Additional kwargs passed to from_var_dict().
+        """
+        loader = LoadExperimentConfig(csv_input) if not isinstance(csv_input, pd.DataFrame) else LoadExperimentConfig(pd.DataFrame(csv_input))
+        var_dict = loader.var_dict
+        return cls.from_var_dict(var_dict, date_str=date_str, config_source=config_source, defaults_dir=defaults_dir, defaults_map=defaults_map, **kwargs)
+
+    # -------------------------
+    # validation & serialization
+    # -------------------------
+    def _validate_hmm_features_structure(hfs: dict) -> List[str]:
+        errs = []
+        if not isinstance(hfs, dict):
+            errs.append("hmm_feature_sets must be a mapping if provided.")
+            return errs
+        for g, info in hfs.items():
+            if not isinstance(info, dict):
+                errs.append(f"hmm_feature_sets['{g}'] must be a mapping with 'features' and 'state'.")
+                continue
+            feats = info.get("features")
+            if not isinstance(feats, dict) or len(feats) == 0:
+                errs.append(f"hmm_feature_sets['{g}'] must include non-empty 'features' mapping.")
+                continue
+            for fname, rng in feats.items():
+                try:
+                    lo, hi = float(rng[0]), float(rng[1])
+                    if lo < 0 or hi <= lo:
+                        errs.append(f"Feature range for {g}:{fname} must satisfy 0 <= lo < hi; got {rng}.")
+                except Exception:
+                    errs.append(f"Feature range for {g}:{fname} is invalid: {rng}")
+        return errs
+
+    def validate(self, require_paths: bool = True, raise_on_error: bool = True) -> List[str]:
+        """
+        Validate the config. If require_paths True, check paths (input_data_path, fasta) exist;
+        attempt to create output_directory if missing.
+        Returns a list of error messages (empty if none). Raises ValueError if raise_on_error True.
+        """
+        errors: List[str] = []
+        if not self.input_data_path:
+            errors.append("input_data_path is required but missing.")
+        if not self.output_directory:
+            errors.append("output_directory is required but missing.")
+        if not self.fasta:
+            errors.append("fasta (reference FASTA) is required but missing.")
+
+        if require_paths:
+            if self.input_data_path and not Path(self.input_data_path).exists():
+                errors.append(f"input_data_path does not exist: {self.input_data_path}")
+            if self.fasta and not Path(self.fasta).exists():
+                errors.append(f"fasta does not exist: {self.fasta}")
+            outp = Path(self.output_directory) if self.output_directory else None
+            if outp and not outp.exists():
+                try:
+                    outp.mkdir(parents=True, exist_ok=True)
+                except Exception as e:
+                    errors.append(f"Could not create output_directory {self.output_directory}: {e}")
+
+        if not (0.0 <= float(self.mapping_threshold) <= 1.0):
+            errors.append("mapping_threshold must be in [0,1].")
+        for t in (self.filter_threshold, self.m6A_threshold, self.m5C_threshold, self.hm5C_threshold):
+            if not (0.0 <= float(t) <= 1.0):
+                errors.append(f"threshold value {t} must be in [0,1].")
+
+        if raise_on_error and errors:
+            raise ValueError("ExperimentConfig validation failed:\n  " + "\n  ".join(errors))
+        
+        errs = _validate_hmm_features_structure(self.hmm_feature_sets)
+        errors.extend(errs)
+
+        return errors
+
+    def to_dict(self) -> Dict[str, Any]:
+        return asdict(self)
+
+    def to_yaml(self, path: Optional[Union[str, Path]] = None) -> str:
+        """
+        Dump config to YAML (string if path None) or save to file at path.
+        If pyyaml is not installed, fallback to JSON for file write.
+        """
+        data = self.to_dict()
+        if path is None:
+            if yaml is None:
+                return json.dumps(data, indent=2)
+            return yaml.safe_dump(data, sort_keys=False)
+        else:
+            p = Path(path)
+            if yaml is None:
+                p.write_text(json.dumps(data, indent=2), encoding="utf8")
+            else:
+                p.write_text(yaml.safe_dump(data, sort_keys=False), encoding="utf8")
+            return str(p)
+
+    def save(self, path: Union[str, Path]) -> str:
+        return self.to_yaml(path)
+
+    def __repr__(self) -> str:
+        return f"<ExperimentConfig modality={self.smf_modality} experiment_name={self.experiment_name} source={self.config_source}>"