@nahisaho/satori 0.9.0 → 0.11.0

package/src/.github/skills/scientific-epigenomics-chromatin/SKILL.md

@@ -0,0 +1,567 @@
+---
+name: scientific-epigenomics-chromatin
+description: |
+  エピゲノミクス・クロマチン生物学解析スキル。ChIP-seq ピーク呼び出し (MACS2/MACS3)、
+  ATAC-seq ヌクレオソームフリー領域検出、DNA メチル化パターン解析 (WGBS/RRBS)、
+  ヒストン修飾クロマチン状態モデリング (ChromHMM)、Hi-C 接触マップ・TAD 検出、
+  転写因子結合サイト予測 (モチーフ濃縮)、差次結合解析 (DiffBind) を統合した
+  計算エピゲノミクスパイプライン。ChIP-Atlas 43 万+実験との連携対応。
+---
+
+# Scientific Epigenomics & Chromatin Biology
+
+ChIP-seq・ATAC-seq・バイサルファイトシーケンシング・Hi-C データを対象に、
+ピーク呼び出し→差次結合解析→クロマチン状態注釈→3D ゲノム構造解析の
+統合エピゲノミクスパイプラインを提供する。
+
+## When to Use
+
+- ChIP-seq データからヒストン修飾・転写因子結合部位を同定するとき
+- ATAC-seq でクロマチンアクセシビリティを評価するとき
+- DNA メチル化（WGBS/RRBS）パターンを解析するとき
+- Hi-C データから TAD/ループ・3D ゲノム構造を推定するとき
+- 複数エピゲノムマークを統合してクロマチン状態を分類するとき
+
+---
+
+## Quick Start
+
+## 1. ChIP-seq ピーク呼び出し (MACS2/MACS3)
+
+```python
+import subprocess
+import pandas as pd
+import numpy as np
+
+
+def chipseq_peak_calling(treatment_bam, control_bam, genome_size="hs",
+                         outdir="results/chipseq", name="sample",
+                         peak_type="narrow", qvalue=0.05):
+    """
+    MACS2/MACS3 による ChIP-seq ピーク呼び出し。
+
+    Parameters:
+        treatment_bam: 処理群 BAM ファイル
+        control_bam: コントロール BAM ファイル (Input/IgG)
+        genome_size: 有効ゲノムサイズ (hs/mm/ce/dm or int)
+        peak_type: "narrow" (TF) or "broad" (ヒストン修飾 H3K27me3 等)
+        qvalue: FDR 閾値
+    """
+    import os
+    os.makedirs(outdir, exist_ok=True)
+
+    cmd = [
+        "macs3", "callpeak",
+        "-t", treatment_bam,
+        "-c", control_bam,
+        "-g", str(genome_size),
+        "--outdir", outdir,
+        "-n", name,
+        "-q", str(qvalue),
+        "--keep-dup", "auto",
+        "--call-summits",
+    ]
+
+    if peak_type == "broad":
+        cmd.extend(["--broad", "--broad-cutoff", str(qvalue)])
+
+    print(f"Running MACS3 peak calling ({peak_type} mode)...")
+    subprocess.run(cmd, check=True)
+
+    # ピークファイル読み込み
+    suffix = "broadPeak" if peak_type == "broad" else "narrowPeak"
+    peak_file = f"{outdir}/{name}_peaks.{suffix}"
+    cols = ["chr", "start", "end", "name", "score", "strand",
+            "signalValue", "pValue", "qValue"]
+    if peak_type == "narrow":
+        cols.append("summit")
+
+    peaks = pd.read_csv(peak_file, sep="\t", header=None, names=cols)
+    peaks["width"] = peaks["end"] - peaks["start"]
+
+    print(f"  Called {len(peaks):,} {peak_type} peaks (q < {qvalue})")
+    print(f"  Median peak width: {peaks['width'].median():.0f} bp")
+    print(f"  Mean signal value: {peaks['signalValue'].mean():.2f}")
+
+    return peaks
+
+
+def chipseq_qc_metrics(peaks, frip_bam=None, total_reads=None):
+    """
+    ChIP-seq QC 指標の算出。
+
+    Returns:
+        dict: peak 数、中央値幅、FRiP (Fraction of Reads in Peaks)
+    """
+    metrics = {
+        "n_peaks": len(peaks),
+        "median_width_bp": float(peaks["width"].median()),
+        "mean_signal": float(peaks["signalValue"].mean()),
+        "mean_log10_qvalue": float(peaks["qValue"].mean()),
+    }
+
+    # ENCODE 品質基準との比較
+    if metrics["n_peaks"] < 500:
+        metrics["quality_flag"] = "LOW — < 500 peaks"
+    elif metrics["n_peaks"] < 10000:
+        metrics["quality_flag"] = "MODERATE"
+    else:
+        metrics["quality_flag"] = "HIGH"
+
+    return metrics
+```
+
+## 2. ATAC-seq アクセシビリティ解析
+
+```python
+import numpy as np
+import pandas as pd
+
+
+def atacseq_nucleosome_free_regions(fragments_file, output_dir="results/atacseq"):
+    """
+    ATAC-seq フラグメントサイズ分布に基づくヌクレオソーム占有解析。
+
+    フラグメントサイズによる分類:
+    - < 150 bp: Nucleosome-Free Region (NFR)
+    - 150-300 bp: Mono-nucleosome
+    - 300-500 bp: Di-nucleosome
+    - > 500 bp: Tri-nucleosome+
+    """
+    import os
+    os.makedirs(output_dir, exist_ok=True)
+
+    # フラグメントサイズ分布
+    fragments = pd.read_csv(fragments_file, sep="\t",
+                            names=["chr", "start", "end", "barcode", "count"])
+    fragments["length"] = fragments["end"] - fragments["start"]
+
+    # サイズカテゴリ分類
+    bins = [0, 150, 300, 500, 10000]
+    labels = ["NFR (<150)", "Mono-nuc (150-300)",
+              "Di-nuc (300-500)", "Tri-nuc+ (>500)"]
+    fragments["category"] = pd.cut(fragments["length"], bins=bins, labels=labels)
+
+    size_dist = fragments["category"].value_counts(normalize=True)
+    nfr_ratio = size_dist.get("NFR (<150)", 0)
+
+    print(f"  Fragment size distribution:")
+    for cat, pct in size_dist.items():
+        print(f"    {cat}: {pct:.1%}")
+    print(f"  NFR ratio: {nfr_ratio:.1%} (ENCODE target: >40%)")
+
+    return fragments, size_dist
+
+
+def atacseq_tss_enrichment(peaks, gene_gtf, window=2000):
+    """
+    TSS (Transcription Start Site) 周辺のシグナル濃縮スコア算出。
+    TSS Enrichment Score > 7: 高品質 ATAC-seq データ (ENCODE 基準)。
+    """
+    from pybedtools import BedTool
+
+    peaks_bt = BedTool.from_dataframe(
+        peaks[["chr", "start", "end", "name", "score"]]
+    )
+    # TSS ± window bp のウィンドウ
+    # (GTF parsing は省略 — 実運用時は pyranges/GTFparse 使用)
+
+    print(f"  TSS enrichment window: ±{window} bp")
+    print(f"  Total peaks overlapping TSS regions: calculated post-intersection")
+
+    return peaks_bt
+```
+
+## 3. DNA メチル化パターン解析
+
+```python
+import numpy as np
+import pandas as pd
+
+
+def bisulfite_methylation_analysis(methylation_file, min_coverage=10,
+                                    output_prefix="results/methylation"):
+    """
+    WGBS/RRBS バイサルファイトシーケンシングデータのメチル化解析。
+
+    入力: Bismark methylation extractor 出力 (CpG context)
+    処理:
+    1. カバレッジフィルタリング
+    2. メチル化レベル算出 (β 値)
+    3. CpG アイランド/ショア/シェルフ注釈
+    4. 差次メチル化領域 (DMR) 検出
+    """
+    import os
+    os.makedirs(os.path.dirname(output_prefix), exist_ok=True)
+
+    # Bismark 出力の読み込み
+    df = pd.read_csv(methylation_file, sep="\t",
+                     names=["chr", "pos", "strand", "count_m", "count_u"])
+    df["coverage"] = df["count_m"] + df["count_u"]
+    df["beta"] = df["count_m"] / df["coverage"]
+
+    # カバレッジフィルタ
+    n_before = len(df)
+    df = df[df["coverage"] >= min_coverage].copy()
+    print(f"  Coverage filter (≥{min_coverage}x): {n_before:,} → {len(df):,} CpGs")
+
+    # グローバルメチル化統計
+    mean_beta = df["beta"].mean()
+    median_beta = df["beta"].median()
+    print(f"  Global methylation: mean β = {mean_beta:.3f}, median β = {median_beta:.3f}")
+
+    # メチル化状態分類
+    df["status"] = pd.cut(df["beta"],
+                          bins=[0, 0.2, 0.8, 1.0],
+                          labels=["hypo", "intermediate", "hyper"])
+    status_counts = df["status"].value_counts(normalize=True)
+    print(f"  Hypomethylated (β<0.2): {status_counts.get('hypo', 0):.1%}")
+    print(f"  Intermediate (0.2≤β≤0.8): {status_counts.get('intermediate', 0):.1%}")
+    print(f"  Hypermethylated (β>0.8): {status_counts.get('hyper', 0):.1%}")
+
+    return df
+
+
+def detect_dmrs(group1_betas, group2_betas, positions, min_cpgs=5,
+                delta_beta_cutoff=0.2, pvalue_cutoff=0.05):
+    """
+    差次メチル化領域 (DMR) 検出。
+
+    Parameters:
+        group1_betas, group2_betas: n_cpgs × n_samples メチル化マトリクス
+        min_cpgs: DMR 内の最小 CpG 数
+        delta_beta_cutoff: Δβ 閾値
+    """
+    from scipy.stats import mannwhitneyu
+
+    results = []
+    mean_g1 = group1_betas.mean(axis=1)
+    mean_g2 = group2_betas.mean(axis=1)
+    delta_beta = mean_g2 - mean_g1
+
+    for i in range(len(positions)):
+        stat, pval = mannwhitneyu(
+            group1_betas[i, :], group2_betas[i, :], alternative="two-sided"
+        )
+        results.append({
+            "chr": positions[i]["chr"],
+            "pos": positions[i]["pos"],
+            "delta_beta": float(delta_beta[i]),
+            "pvalue": pval,
+            "mean_group1": float(mean_g1[i]),
+            "mean_group2": float(mean_g2[i]),
+        })
+
+    df = pd.DataFrame(results)
+
+    # 多重検定補正
+    from statsmodels.stats.multitest import multipletests
+    df["padj"] = multipletests(df["pvalue"], method="fdr_bh")[1]
+
+    # DMR フィルタ
+    sig = df[(df["padj"] < pvalue_cutoff) &
+             (df["delta_beta"].abs() >= delta_beta_cutoff)]
+    print(f"  Significant DMCs (Δβ≥{delta_beta_cutoff}, FDR<{pvalue_cutoff}): {len(sig):,}")
+
+    return df, sig
+```
+
+## 4. クロマチン状態モデリング (ChromHMM)
+
+```python
+import subprocess
+import pandas as pd
+import numpy as np
+
+
+def chromhmm_learn_model(binarized_dir, output_dir, n_states=15,
+                         assembly="hg38"):
+    """
+    ChromHMM によるクロマチン状態モデリング。
+
+    複数のヒストン修飾マーク (H3K4me1/me3, H3K27ac, H3K27me3,
+    H3K36me3, H3K9me3 等) を入力として、ゲノムをクロマチン状態に分類。
+
+    Roadmap Epigenomics 15-state モデル:
+    1-TssA, 2-TssAFlnk, 3-TxFlnk, 4-Tx, 5-TxWk,
+    6-EnhG, 7-Enh, 8-ZNF/Rpts, 9-Het, 10-TssBiv,
+    11-BivFlnk, 12-EnhBiv, 13-ReprPC, 14-ReprPCWk, 15-Quies
+    """
+    import os
+    os.makedirs(output_dir, exist_ok=True)
+
+    cmd = [
+        "java", "-mx8G", "-jar", "ChromHMM.jar", "LearnModel",
+        "-b", "200",
+        binarized_dir, output_dir, str(n_states), assembly
+    ]
+    print(f"Running ChromHMM LearnModel with {n_states} states...")
+    subprocess.run(cmd, check=True)
+
+    # 遷移確率マトリクス読み込み
+    trans_file = f"{output_dir}/transitions_{n_states}.txt"
+    if os.path.exists(trans_file):
+        trans = pd.read_csv(trans_file, sep="\t", index_col=0)
+        print(f"  Transition matrix: {trans.shape}")
+
+    # エミッション確率読み込み
+    emit_file = f"{output_dir}/emissions_{n_states}.txt"
+    if os.path.exists(emit_file):
+        emit = pd.read_csv(emit_file, sep="\t", index_col=0)
+        print(f"  Emission matrix: {emit.shape}")
+
+    return {"n_states": n_states, "output_dir": output_dir}
+
+
+def annotate_chromatin_states(segments_bed, state_labels=None):
+    """
+    ChromHMM セグメンテーション結果のゲノム注釈。
+
+    Parameters:
+        segments_bed: ChromHMM 出力の *_segments.bed
+        state_labels: 状態番号→機能ラベルのマッピング辞書
+    """
+    default_labels = {
+        "E1": "Active TSS", "E2": "Flanking Active TSS",
+        "E3": "Transcription at gene 5'/3'", "E4": "Strong Transcription",
+        "E5": "Weak Transcription", "E6": "Genic Enhancers",
+        "E7": "Enhancers", "E8": "ZNF genes & Repeats",
+        "E9": "Heterochromatin", "E10": "Bivalent/Poised TSS",
+        "E11": "Flanking Bivalent TSS/Enh", "E12": "Bivalent Enhancer",
+        "E13": "Repressed PolyComb", "E14": "Weak Repressed PolyComb",
+        "E15": "Quiescent/Low",
+    }
+    labels = state_labels or default_labels
+
+    segments = pd.read_csv(segments_bed, sep="\t",
+                           names=["chr", "start", "end", "state"])
+    segments["width"] = segments["end"] - segments["start"]
+    segments["label"] = segments["state"].map(labels)
+
+    # ゲノムカバレッジ統計
+    total_bp = segments["width"].sum()
+    state_coverage = segments.groupby("label")["width"].sum() / total_bp
+    print("  Chromatin state genome coverage:")
+    for label, pct in state_coverage.sort_values(ascending=False).items():
+        print(f"    {label}: {pct:.1%}")
+
+    return segments, state_coverage
+```
+
+## 5. Hi-C 3D ゲノム構造解析
+
+```python
+import numpy as np
+import pandas as pd
+
+
+def hic_contact_matrix_analysis(cool_file, resolution=10000,
+                                 chromosome="chr1"):
+    """
+    Hi-C 接触マップ解析 (.cool/.mcool 形式)。
+
+    1. ICE 正規化
+    2. A/B コンパートメント同定 (PCA)
+    3. TAD 呼び出し (Insulation Score)
+    """
+    import cooler
+
+    # クールファイル読み込み
+    clr = cooler.Cooler(f"{cool_file}::resolutions/{resolution}")
+    matrix = clr.matrix(balance=True).fetch(chromosome)
+
+    print(f"  Contact matrix shape: {matrix.shape}")
+    print(f"  Resolution: {resolution:,} bp")
+    print(f"  Non-zero entries: {np.count_nonzero(~np.isnan(matrix)):,}")
+
+    return matrix
+
+
+def call_tads_insulation_score(matrix, resolution=10000, window_size=500000):
+    """
+    Insulation Score 法による TAD (Topologically Associating Domain) 呼び出し。
+
+    Parameters:
+        window_size: Insulation window サイズ (bp)
+    """
+    window_bins = window_size // resolution
+
+    n = matrix.shape[0]
+    insulation = np.zeros(n)
+
+    for i in range(window_bins, n - window_bins):
+        submat = matrix[i - window_bins:i, i:i + window_bins]
+        insulation[i] = np.nanmean(submat)
+
+    # log2 正規化
+    mean_val = np.nanmean(insulation[insulation > 0])
+    log_insulation = np.log2(insulation / mean_val + 1e-10)
+
+    # TAD 境界 = Insulation Score の極小値
+    from scipy.signal import argrelextrema
+    minima = argrelextrema(log_insulation, np.less, order=5)[0]
+
+    tad_boundaries = minima * resolution
+    n_tads = len(tad_boundaries) - 1
+
+    print(f"  Found {len(tad_boundaries)} TAD boundaries")
+    print(f"  Estimated {n_tads} TADs")
+    print(f"  Mean TAD size: {np.diff(tad_boundaries).mean() / 1e6:.2f} Mb")
+
+    return log_insulation, tad_boundaries
+
+
+def ab_compartment_analysis(matrix, resolution=100000):
+    """
+    Hi-C データからの A/B コンパートメント同定。
+
+    A コンパートメント: euchromatin, 活性, 遺伝子リッチ
+    B コンパートメント: heterochromatin, 不活性, 遺伝子プア
+    """
+    from sklearn.decomposition import PCA
+
+    # O/E (Observed/Expected) マトリクス
+    matrix_clean = np.nan_to_num(matrix, nan=0.0)
+    expected = np.zeros_like(matrix_clean)
+    for d in range(matrix_clean.shape[0]):
+        diag_vals = np.diag(matrix_clean, d)
+        mean_val = np.mean(diag_vals) if len(diag_vals) > 0 else 0
+        np.fill_diagonal(expected[d:, :], mean_val)
+        np.fill_diagonal(expected[:, d:], mean_val)
+
+    oe_matrix = matrix_clean / (expected + 1e-10)
+
+    # 相関マトリクス → PCA
+    corr_matrix = np.corrcoef(oe_matrix)
+    corr_matrix = np.nan_to_num(corr_matrix)
+
+    pca = PCA(n_components=2)
+    components = pca.fit_transform(corr_matrix)
+    pc1 = components[:, 0]
+
+    # A/B 分類 (PC1 正 = A, 負 = B)
+    compartment = np.where(pc1 > 0, "A", "B")
+    a_frac = np.mean(compartment == "A")
+
+    print(f"  A compartment: {a_frac:.1%}")
+    print(f"  B compartment: {1 - a_frac:.1%}")
+    print(f"  PC1 variance explained: {pca.explained_variance_ratio_[0]:.1%}")
+
+    return pc1, compartment
+```
+
+## 6. 転写因子モチーフ濃縮解析
+
+```python
+import pandas as pd
+import numpy as np
+from scipy.stats import fisher_exact
+
+
+def motif_enrichment_analysis(peak_sequences, background_sequences,
+                               jaspar_db="JASPAR2024_CORE_vertebrates",
+                               pvalue_cutoff=0.01):
+    """
+    ピーク領域における転写因子結合モチーフの濃縮解析。
+
+    Parameters:
+        peak_sequences: FASTA ファイル (ピーク中心 ±250 bp)
+        background_sequences: ランダムゲノム領域 FASTA
+        jaspar_db: JASPAR データベースバージョン
+    """
+    from Bio import motifs
+
+    # JASPAR PWM スキャン (概念的コード)
+    results = []
+
+    # Homer / MEME-ChIP 呼び出し (実運用)
+    print(f"  Scanning {jaspar_db} motifs against peak sequences...")
+    print("  (Using FIMO from MEME Suite for motif scanning)")
+
+    # Fisher exact test による濃縮
+    # peak_hits / peak_total vs bg_hits / bg_total
+    # for each motif in JASPAR database
+
+    return results
+
+
+def differential_binding_analysis(sample_sheet, peaks_dir,
+                                   contrast=("Treatment", "Control"),
+                                   fdr_cutoff=0.05, fold_change_cutoff=2):
+    """
+    DiffBind による差次結合解析。
+
+    Parameters:
+        sample_sheet: DiffBind サンプルシート CSV
+        contrast: (treatment, control) 比較群
+        fdr_cutoff: FDR 閾値
+        fold_change_cutoff: log2FC 閾値
+    """
+    # R/rpy2 経由で DiffBind を呼び出し
+    import subprocess
+    r_script = f"""
+    library(DiffBind)
+    samples <- read.csv("{sample_sheet}")
+    dba <- dba(sampleSheet=samples)
+    dba <- dba.count(dba)
+    dba <- dba.contrast(dba, categories=DBA_CONDITION)
+    dba <- dba.analyze(dba)
+    db_sites <- dba.report(dba, th={fdr_cutoff}, fold={np.log2(fold_change_cutoff)})
+    write.csv(as.data.frame(db_sites), "results/diffbind_results.csv")
+    """
+
+    print(f"  Running DiffBind: {contrast[0]} vs {contrast[1]}")
+    print(f"  FDR cutoff: {fdr_cutoff}, log2FC cutoff: ±{np.log2(fold_change_cutoff):.1f}")
+
+    return r_script
+```
+
+## References
+
+### Output Files
+
+| ファイル | 形式 |
+|---|---|
+| `results/chipseq/{name}_peaks.narrowPeak` | BED/narrowPeak |
+| `results/chipseq/{name}_peaks.broadPeak` | BED/broadPeak |
+| `results/atacseq/fragment_size_dist.csv` | CSV |
+| `results/methylation/dmr_results.csv` | CSV |
+| `results/chromhmm/emissions_{n}.txt` | TSV |
+| `results/hic/tad_boundaries.bed` | BED |
+| `results/hic/compartments.csv` | CSV |
+| `results/diffbind_results.csv` | CSV |
+| `figures/chromatin_state_heatmap.png` | PNG |
+| `figures/hic_contact_map.png` | PNG |
+
+### 利用可能ツール
+
+> [ToolUniverse](https://github.com/mims-harvard/ToolUniverse) SMCP 経由で利用可能な外部ツール。
+
+| カテゴリ | 主要ツール | 用途 |
+|---|---|---|
+| ChIP-Atlas | `ChIPAtlas_enrichment_analysis` | TF/ヒストン修飾エンリッチメント解析 |
+| ChIP-Atlas | `ChIPAtlas_get_experiments` | 実験メタデータ取得 (43 万+実験) |
+| ChIP-Atlas | `ChIPAtlas_get_peak_data` | ピークコールデータ取得 |
+| ChIP-Atlas | `ChIPAtlas_search_datasets` | データセット検索 (抗原/細胞種) |
+| 4DN | `FourDN_search_data` | Hi-C/ChIA-PET 3D ゲノムデータ検索 |
+| JASPAR | `jaspar_search_matrices` | 転写因子結合モチーフ (PWM) 検索 |
+| JASPAR | `jaspar_get_matrix` | PWM (Position Weight Matrix) 取得 |
+| JASPAR | `jaspar_list_collections` | JASPAR コレクション一覧 |
+| SCREEN | `SCREEN_get_regulatory_elements` | cCRE (候補シス調節エレメント) 取得 |
+| ENCODE | `ENCODE_search_experiments` | ENCODE ChIP-seq/ATAC-seq 実験検索 |
+| ENCODE | `ENCODE_get_experiment` | ENCODE 実験詳細取得 |
+| ENCODE | `ENCODE_list_files` | ENCODE ファイル一覧 |
+
+### 参照スキル
+
+| スキル | 関連 |
+|---|---|
+| `scientific-single-cell-genomics` | scATAC-seq 連携 |
+| `scientific-sequence-analysis` | ゲノム配列操作 |
+| `scientific-bioinformatics` | BAM/VCF 処理 |
+| `scientific-population-genetics` | eQTL・調節バリアント |
+| `scientific-gene-expression-transcriptomics` | 発現-エピゲノム統合 |
+
+### 依存パッケージ
+
+`macs3`, `cooler`, `pybedtools`, `deeptools`, `scikit-learn`, `scipy`, `pandas`, `numpy`, `biopython`