gsMap3D 0.1.0a1__py3-none-any.whl

gsMap/utils/generate_r2_matrix.py

@@ -0,0 +1,610 @@
+"""
+Module for reading and processing PLINK genotype data and calculating LD scores.
+
+Note:
+This code is adapted and modified from:
+https://github.com/bulik/ldsc/blob/master/ldsc/ldscore.py
+"""
+
+import logging
+
+import bitarray as ba
+import numba
+import numpy as np
+import pandas as pd
+import pyranges as pr
+import torch
+from tqdm import tqdm
+
+from gsMap.utils.torch_utils import torch_device, torch_sync
+
+# Configure logger
+logger = logging.getLogger("gsMap.utils.plink_ldscore_tool")
+
+
+@numba.njit
+def getBlockLefts(coords: np.ndarray, max_dist: float):
+    """
+    Converts coordinates + max block length to a list of coordinates of the leftmost
+    SNPs to be included in blocks.
+    """
+    M = len(coords)
+    j = 0
+    block_left = np.zeros(M)
+    for i in range(M):
+        while j < M and abs(coords[j] - coords[i]) > max_dist:
+            j += 1
+
+        block_left[i] = j
+    return block_left
+
+
+@numba.njit
+def normalized_snps(X: np.ndarray, b: int, minorRef, freq, currentSNP):
+    """
+    Normalize the SNPs and impute the missing ones with the mean
+
+    Parameters
+    ----------
+    fam_file : str
+        Path to the FAM file
+
+    Returns
+    -------
+    pd.DataFrame
+        DataFrame containing FAM data
+    """
+    Y = np.zeros(X.shape, dtype="float32")
+
+    for j in range(0, b):
+        newsnp = X[:, j]
+        ii = newsnp != 9
+        avg = np.mean(newsnp[ii])
+        newsnp[np.logical_not(ii)] = avg
+        denom = np.std(newsnp)
+        if denom == 0:
+            denom = 1
+
+        if minorRef is not None and freq[currentSNP + j] > 0.5:
+            denom = denom * -1
+
+        Y[:, j] = (newsnp - avg) / denom
+    return Y
+
+
+def l2_unbiased(x: torch.Tensor, n: int):
+    """
+    Calculate the unbiased estimate of L2.
+    """
+    denom = n - 2 if n > 2 else n  # allow n<2 for testing purposes
+    sq = torch.square(x)
+    return sq - (1 - sq) / denom
+
+
+class PlinkBEDFile:
+    """
+    Interface for Plink .bed format for reading and processing genotype data.
+    """
+
+    def __init__(self, bfile_prefix):
+        """
+        Initialize the PlinkBEDFile from a PLINK file prefix.
+
+        Parameters
+        ----------
+        bfile_prefix : str
+            PLINK file prefix (without .bed/.bim/.fam extension)
+        """
+        # Initialize bitarray for bed code mapping
+        self._bedcode = {
+            2: ba.bitarray("11"),
+            9: ba.bitarray("10"),
+            1: ba.bitarray("01"),
+            0: ba.bitarray("00"),
+        }
+
+        # Load BIM file
+        self.bim_df = self.load_bim(f"{bfile_prefix}.bim")
+
+        # Load FAM file
+        self.fam_df = self.load_fam(f"{bfile_prefix}.fam")
+
+        # Set up initial parameters
+        self.m_original = len(self.bim_df)
+        self.n_original = len(self.fam_df)
+
+        # Read the bed file
+        logger.info(f"Loading Plink genotype data from {bfile_prefix}.bed")
+        (self.nru_original, self.geno_original) = self._read(
+            f"{bfile_prefix}.bed", self.m_original, self.n_original
+        )
+
+        # Pre-calculate MAF for all SNPs
+        logger.info("Calculating MAF and QC for all SNPs")
+        self.all_snp_info = self._calculate_all_snp_info()
+
+        # Filter out invalid SNPs
+        valid_mask = self.all_snp_info["valid_snp"]
+        if num_invalid := np.sum(~valid_mask):
+            logger.warning(
+                f"Filtering out {num_invalid} bad quality SNPs: {self.bim_df.loc[~valid_mask, 'SNP'].tolist()}"
+            )
+        else:
+            logger.info("All SNPs passed the basic quality check")
+
+        # Create new genotype data with only the valid SNPs
+        new_geno = ba.bitarray()
+        for j in np.arange(self.m_original)[valid_mask]:
+            new_geno += self.geno_original[
+                2 * self.nru_original * j : 2 * self.nru_original * (j + 1)
+            ]
+
+        # Update original data to only include valid SNPs
+        self.geno_original = new_geno
+
+        # Only keep valid SNPs
+        self.bim_df = self.bim_df.loc[valid_mask].reset_index(drop=True)
+        self.m_original = len(self.bim_df)
+        self.kept_snps = np.arange(self.m_original)
+
+        # Initialize current state variables
+        self._currentSNP = 0
+        self.m = self.m_original
+        self.n = self.n_original
+        self.nru = self.nru_original
+        self.geno = self.geno_original.copy()
+
+        # Update frequency info based on valid SNPs
+        self.freq = self.all_snp_info["freq"][valid_mask]
+        self.maf = np.minimum(self.freq, 1 - self.freq)
+        self.sqrtpq = np.sqrt(self.freq * (1 - self.freq))
+
+        # Add MAF to the BIM dataframe
+        self.bim_df["MAF"] = self.maf
+
+        logger.info(f"Loaded genotype data with {self.m} SNPs and {self.n} individuals")
+
+    @staticmethod
+    def load_bim(bim_file):
+        """
+        Load a BIM file into a pandas DataFrame.
+
+        Parameters
+        ----------
+        bim_file : str
+            Path to the BIM file
+
+        Returns
+        -------
+        pd.DataFrame
+            DataFrame containing BIM data
+        """
+        df = pd.read_csv(
+            bim_file, sep="\t", header=None, names=["CHR", "SNP", "CM", "BP", "A1", "A2"]
+        )
+        return df
+
+    @staticmethod
+    def convert_bim_to_pyrange(bim_df) -> pr.PyRanges:
+        bim_pr = bim_df.copy()
+        bim_pr.drop(columns=["MAF"], inplace=True)
+        bim_pr.columns = ["Chromosome", "SNP", "CM", "Start", "A1", "A2"]
+        bim_pr.Chromosome = "chr" + bim_pr["Chromosome"].astype(str)
+
+        # Adjust coordinates (BIM is 1-based, PyRanges uses 0-based)
+        bim_pr["End"] = bim_pr["Start"].copy()
+        bim_pr["Start"] = bim_pr["Start"] - 1
+
+        bim_pr = pr.PyRanges(bim_pr)
+
+        return bim_pr
+
+    @staticmethod
+    def load_fam(fam_file):
+        """
+        Load a FAM file into a pandas DataFrame.
+
+        Parameters
+        ----------
+        fam_file : str
+            Path to the FAM file
+
+        Returns
+        -------
+        pd.DataFrame
+            DataFrame containing FAM data
+        """
+        df = pd.read_csv(fam_file, sep=r"\s+", header=None, usecols=[1], names=["IID"])
+        return df
+
+    def _read(self, fname, m, n):
+        """
+        Read the bed file and return the genotype data.
+        """
+        if not fname.endswith(".bed"):
+            raise ValueError(".bed filename must end in .bed")
+
+        fh = open(fname, "rb")
+        magicNumber = ba.bitarray(endian="little")
+        magicNumber.fromfile(fh, 2)
+        bedMode = ba.bitarray(endian="little")
+        bedMode.fromfile(fh, 1)
+        e = (4 - n % 4) if n % 4 != 0 else 0
+        nru = n + e
+
+        # Check magic number
+        if magicNumber != ba.bitarray("0011011011011000"):
+            raise OSError("Magic number from Plink .bed file not recognized")
+
+        if bedMode != ba.bitarray("10000000"):
+            raise OSError("Plink .bed file must be in default SNP-major mode")
+
+        # Check file length
+        geno = ba.bitarray(endian="little")
+        geno.fromfile(fh)
+        self._test_length(geno, m, nru)
+        return (nru, geno)
+
+    def _test_length(self, geno, m, nru):
+        """
+        Test if the genotype data has the expected length.
+        """
+        exp_len = 2 * m * nru
+        real_len = len(geno)
+        if real_len != exp_len:
+            s = "Plink .bed file has {n1} bits, expected {n2}"
+            raise OSError(s.format(n1=real_len, n2=exp_len))
+
+    def _calculate_all_snp_info(self):
+        """
+        Pre-calculate MAF and other information for all SNPs.
+
+        Returns
+        -------
+        dict
+            Dictionary containing information for all SNPs
+        """
+        nru = self.nru_original
+        n = self.n_original
+        m = self.m_original
+        geno = self.geno_original
+
+        snp_info = {
+            "freq": np.zeros(m),  # Allele frequencies
+            "het_miss_count": np.zeros(m),  # Count of het or missing genotypes
+            "valid_snp": np.zeros(m, dtype=bool),  # Whether SNP passes basic criteria
+        }
+
+        # For each SNP, calculate statistics
+        for j in range(m):
+            z = geno[2 * nru * j : 2 * nru * (j + 1)]
+            A = z[0::2]
+            a = A.count()
+            B = z[1::2]
+            b = B.count()
+            c = (A & B).count()
+            major_ct = b + c  # number of copies of the major allele
+            n_nomiss = n - a + c  # number of individuals with nonmissing genotypes
+            f = major_ct / (2 * n_nomiss) if n_nomiss > 0 else 0
+            het_miss_ct = a + b - 2 * c  # count of SNPs that are het or missing
+
+            snp_info["freq"][j] = f
+            snp_info["het_miss_count"][j] = het_miss_ct
+            snp_info["valid_snp"][j] = het_miss_ct < n  # Basic validity check
+
+        return snp_info
+
+    def apply_filters(self, keep_snps=None, keep_indivs=None, mafMin=None):
+        """
+        Apply filters to the genotype data without reloading the bed file.
+
+        Parameters
+        ----------
+        keep_snps : array-like, optional
+            Indices of SNPs to keep.
+        keep_indivs : array-like, optional
+            Indices of individuals to keep.
+        mafMin : float, optional
+            Minimum minor allele frequency.
+
+        Returns
+        -------
+        self
+            Returns self for method chaining.
+        """
+        # Reset to original state first
+        self.geno = self.geno_original.copy()
+        self.m = self.m_original
+        self.n = self.n_original
+        self.nru = self.nru_original
+        self._currentSNP = 0
+
+        # Initialize with all SNPs
+        kept_snps = np.arange(self.m_original)
+
+        # Apply MAF filter using pre-calculated values
+        if mafMin is not None and mafMin > 0:
+            # Remove the redundant valid_snp check since all SNPs are already valid
+            maf_mask = self.maf > mafMin
+            kept_snps = kept_snps[maf_mask]
+            logger.info(f"After MAF filtering (>{mafMin}), {len(kept_snps)} SNPs remain")
+
+        # Apply SNP filter if specified
+        if keep_snps is not None:
+            keep_snps = np.array(keep_snps, dtype="int")
+            if np.any(keep_snps > self.m_original):
+                raise ValueError("keep_snps indices out of bounds")
+
+            # Intersect with current kept_snps
+            kept_snps = np.intersect1d(kept_snps, keep_snps)
+            logger.info(f"After keep_snps filtering, {len(kept_snps)} SNPs remain")
+
+        # Filter SNPs in the genotype data
+        if len(kept_snps) < self.m_original:
+            # Create new genotype data with only the kept SNPs
+            new_geno = ba.bitarray()
+            for j in kept_snps:
+                new_geno += self.geno_original[2 * self.nru * j : 2 * self.nru * (j + 1)]
+            self.geno = new_geno
+            self.m = len(kept_snps)
+
+        # Filter individuals if specified
+        if keep_indivs is not None:
+            keep_indivs = np.array(keep_indivs, dtype="int")
+            if np.any(keep_indivs > self.n):
+                raise ValueError("keep_indivs indices out of bounds")
+
+            (self.geno, self.m, self.n) = self._filter_indivs(
+                self.geno, keep_indivs, self.m, self.n
+            )
+
+            if self.n > 0:
+                logger.info(f"After filtering, {self.n} individuals remain")
+            else:
+                raise ValueError("After filtering, no individuals remain")
+
+        # Update kept_snps and other attributes
+        self.kept_snps = kept_snps
+        self.freq = self.all_snp_info["freq"][kept_snps]
+        self.maf = np.minimum(self.freq, 1 - self.freq)
+        self.sqrtpq = np.sqrt(self.freq * (1 - self.freq))
+
+        return self
+
+    def _filter_indivs(self, geno, keep_indivs, m, n):
+        """
+        Filter individuals based on the keep_indivs parameter.
+        """
+        n_new = len(keep_indivs)
+        e = (4 - n_new % 4) if n_new % 4 != 0 else 0
+        nru_new = n_new + e
+        nru = self.nru
+        z = ba.bitarray(m * 2 * nru_new, endian="little")
+        z.setall(0)
+        for e, i in enumerate(keep_indivs):
+            z[2 * e :: 2 * nru_new] = geno[2 * i :: 2 * nru]
+            z[2 * e + 1 :: 2 * nru_new] = geno[2 * i + 1 :: 2 * nru]
+        self.nru = nru_new
+        return (z, m, n_new)
+
+    def get_snps_by_maf(self, mafMin):
+        """
+        Get the list of SNPs that pass the MAF threshold.
+
+        Parameters
+        ----------
+        mafMin : float
+            Minimum MAF threshold
+
+        Returns
+        -------
+        list
+            List of SNP IDs that pass the MAF threshold
+        """
+        maf_mask = self.maf > mafMin
+
+        # Get SNP names from the BIM dataframe
+        snp_pass_maf = self.bim_df.loc[maf_mask, "SNP"].tolist()
+
+        logger.info(f"{len(snp_pass_maf)} SNPs with MAF > f{mafMin}")
+
+        return snp_pass_maf
+
+    def get_ldscore(self, annot_matrix=None, ld_wind=1.0, ld_unit="CM", keep_snps_index=None):
+        """
+        Calculate LD scores using an annotation matrix.
+
+        Parameters
+        ----------
+        annot_matrix : np.ndarray, optional
+            Annotation matrix. If None, uses a matrix of all ones.
+        ld_wind : float, optional
+            LD window size, by default 1.0
+        ld_unit : str, optional
+            Unit for the LD window, by default "CM"
+        keep_snps_index : list[int], optional
+            Indices of SNPs to keep, by default None
+
+        Returns
+        -------
+        np.ndarray
+            Array with calculated LD scores
+        """
+        # Apply filters if needed
+        if keep_snps_index is not None:
+            original_kept_snps = self.kept_snps.copy()
+            self.apply_filters(keep_snps=keep_snps_index)
+
+        # Configure LD window based on specified unit
+        if ld_unit == "SNP":
+            max_dist = ld_wind
+            coords = np.array(range(self.m))
+        elif ld_unit == "KB":
+            max_dist = ld_wind * 1000
+            coords = np.array(self.bim_df.loc[self.kept_snps, "BP"])
+        elif ld_unit == "CM":
+            max_dist = ld_wind
+            coords = np.array(self.bim_df.loc[self.kept_snps, "CM"])
+            # Check if the CM is all 0
+            if np.all(coords == 0):
+                logger.warning(
+                    "All CM values are 0. Using 1MB window size for LD score calculation."
+                )
+                max_dist = 1_000_000
+                coords = np.array(self.bim_df.loc[self.kept_snps, "BP"])
+        else:
+            raise ValueError(f"Invalid ld_wind_unit: {ld_unit}. Must be one of: SNP, KB, CM")
+
+        # Calculate blocks for LD computation
+        block_left = getBlockLefts(coords, max_dist)
+        assert block_left.sum() > 0, "Invalid window size, please check the ld_wind parameter."
+
+        # Calculate LD scores
+        ld_scores = self.ldScoreVarBlocks(block_left, 100, annot=annot_matrix)
+
+        # Restore original state if filters were applied
+        if keep_snps_index is not None:
+            self.apply_filters(keep_snps=original_kept_snps)
+
+        return ld_scores
+
+    def restart(self):
+        """
+        Reset the current SNP index to 0.
+        """
+        self._currentSNP = 0
+
+    def nextSNPs(self, b, minorRef=None):
+        """
+        Unpacks the binary array of genotypes and returns an n x b matrix of floats of
+        normalized genotypes for the next b SNPs.
+        """
+        try:
+            b = int(b)
+            if b <= 0:
+                raise ValueError("b must be > 0")
+        except TypeError as e:
+            raise TypeError("b must be an integer") from e
+
+        if self._currentSNP + b > self.m:
+            s = "{b} SNPs requested, {k} SNPs remain"
+            raise ValueError(s.format(b=b, k=(self.m - self._currentSNP)))
+
+        c = self._currentSNP
+        n = self.n
+        nru = self.nru
+        slice = self.geno[2 * c * nru : 2 * (c + b) * nru]
+        X = np.array(slice.decode(self._bedcode), dtype="float32").reshape((b, nru)).T
+        X = X[0:n, :]
+        Y = normalized_snps(X, b, minorRef, self.freq, self._currentSNP)
+
+        self._currentSNP += b
+        return Y
+
+    def ldScoreVarBlocks(self, block_left: np.ndarray, c, annot=None):
+        """
+        Computes an unbiased estimate of L2(j) for j=1,..,M.
+        """
+
+        def func(x):
+            return l2_unbiased(x, self.n)
+
+        snp_getter = self.nextSNPs
+        return self._corSumVarBlocks(block_left, c, func, snp_getter, annot)
+
+    def _corSumVarBlocks(self, block_left, c, func, snp_getter, annot=None):
+        """
+        Calculate the sum of correlation coefficients.
+        """
+        m, n = self.m, self.n
+        block_sizes = np.array(np.arange(m) - block_left)
+        block_sizes = np.ceil(block_sizes / c) * c
+        if annot is None:
+            annot = np.ones((m, 1), dtype="float32")
+        else:
+            # annot = annot.astype("float32")  # Ensure annot is float32
+            annot_m = annot.shape[0]
+            if annot_m != self.m:
+                raise ValueError("Incorrect number of SNPs in annot")
+
+        n_a = annot.shape[1]  # number of annotations
+        cor_sum = np.zeros((m, n_a), dtype="float32")
+        # b = index of first SNP for which SNP 0 is not included in LD Score
+        b = np.nonzero(block_left > 0)
+        if np.any(b):
+            b = b[0][0]
+        else:
+            b = m
+        b = int(np.ceil(b / c) * c)  # round up to a multiple of c
+        if b > m:
+            c = 1
+            b = m
+
+        l_A = 0  # l_A := index of leftmost SNP in matrix A
+
+        device = torch_device()
+        A = torch.from_numpy(snp_getter(b)).to(device)  # This now returns float32 data
+        cor_sum = torch.from_numpy(cor_sum).to(device)
+        annot = torch.from_numpy(annot).to(device)
+        rfuncAB = torch.zeros((b, c), dtype=torch.float32, device=device)
+        rfuncBB = torch.zeros((c, c), dtype=torch.float32, device=device)
+
+        # chunk inside of block
+        for l_B in np.arange(0, b, c):  # l_B := index of leftmost SNP in matrix B
+            B = A[:, l_B : l_B + c]
+            # ld matrix
+            torch.mm(A.T, B / n, out=rfuncAB)
+            # ld matrix square
+            rfuncAB = func(rfuncAB)
+            cor_sum[l_A : l_A + b, :] += torch.mm(rfuncAB, annot[l_B : l_B + c, :].float())
+
+        # chunk to right of block
+        b0 = b
+        md = int(c * np.floor(m / c))
+        end = md + 1 if md != m else md
+        for l_B in tqdm(np.arange(b0, end, c), desc="Compute SNP Gene Weight"):
+            # check if the annot matrix is all zeros for this block + chunk
+            # this happens w/ sparse categories (i.e., pathways)
+            # update the block
+            old_b = b
+            b = int(block_sizes[l_B])
+            if l_B > b0 and b > 0:
+                # block_size can't increase more than c
+                # block_size can't be less than c unless it is zero
+                # both of these things make sense
+                A = torch.hstack((A[:, old_b - b + c : old_b], B))
+                l_A += old_b - b + c
+            elif l_B == b0 and b > 0:
+                A = A[:, b0 - b : b0]
+                l_A = b0 - b
+            elif b == 0:  # no SNPs to left in window, e.g., after a sequence gap
+                A = torch.zeros((n, 0), dtype=torch.float32, device=device)
+                l_A = l_B
+            if l_B == md:
+                c = m - md
+                rfuncAB = torch.zeros((b, c), dtype=torch.float32, device=device)
+                rfuncBB = torch.zeros((c, c), dtype=torch.float32, device=device)
+            if b != old_b:
+                rfuncAB = torch.zeros((b, c), dtype=torch.float32, device=device)
+
+            B = torch.from_numpy(snp_getter(c)).to(device)  # This now returns float32 data
+
+            annot_l_A = annot[l_A : l_A + b, :].float()
+            annot_l_B = annot[l_B : l_B + c, :].float()
+            p1 = torch.all(annot_l_A == 0)
+            p2 = torch.all(annot_l_B == 0)
+            if p1 and p2:
+                continue
+
+            B_n = B / n
+
+            rfuncAB = func(torch.mm(A.T, B_n))
+            cor_sum[l_A : l_A + b, :] += torch.mm(rfuncAB, annot_l_B)
+            cor_sum[l_B : l_B + c, :] += torch.mm(annot_l_A.T, rfuncAB).T
+
+            rfuncBB = func(torch.mm(B.T, B_n))
+            cor_sum[l_B : l_B + c, :] += torch.mm(rfuncBB, annot_l_B)
+
+        torch_sync()
+
+        return cor_sum.cpu().numpy()