File size: 20,583 Bytes
6cc35b0 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 | """Multiple-sequence-alignment value objects and lossless encodings."""
from __future__ import annotations
import dataclasses
import string
from collections.abc import Sequence
from dataclasses import dataclass
from functools import cached_property
from itertools import islice
from typing import Any
import numpy as np
from Bio import SeqIO
from scipy.spatial.distance import cdist
from .esmfold2_misc import slice_any_object
from .esmfold2_msa_filter_sequences import greedy_select_indices, hhfilter
from .esmfold2_parsing import FastaEntry, read_sequences, write_sequences
from .esmfold2_sequential_dataclass import SequentialDataclass
from .esmfold2_system import PathOrBuffer
_A3M_INSERTION_DELETE_TABLE = str.maketrans(
dict.fromkeys(string.ascii_lowercase + ".")
)
_SERIALIZATION_VERSION = 1
_UINT32_BYTES = 4
def is_a3m_insertion(character: str) -> bool:
"""Return whether a character is an A3M insertion marker."""
return character == "." or character.islower()
def remove_insertions_from_sequence(sequence: str) -> str:
"""Remove lowercase residues and dot insertion markers from an A3M row."""
return sequence.translate(_A3M_INSERTION_DELETE_TABLE)
def a3m_deletion_counts(sequence: str) -> np.ndarray:
"""Count insertions preceding each A3M match column."""
codes = np.frombuffer(sequence.encode("ascii"), dtype=np.uint8)
lowercase = (codes >= ord("a")) & (codes <= ord("z"))
insertion_mask = lowercase | (codes == ord("."))
prefix_counts = np.concatenate(([0], np.cumsum(insertion_mask)))
match_positions = np.flatnonzero(~insertion_mask)
return np.diff(prefix_counts[match_positions], prepend=0)
def _parse_full_payload(data: bytes) -> tuple[np.ndarray, list[str]]:
version = int.from_bytes(data[:1], "little")
if version != _SERIALIZATION_VERSION:
raise ValueError(f"Unsupported version: {version}")
seqlen = int.from_bytes(data[1:5], "little")
depth = int.from_bytes(data[5:9], "little")
body = data[9:]
split = seqlen * depth
array = np.frombuffer(body[:split], dtype="|S1").reshape(depth, seqlen)
headers = [header for header in body[split:].decode().split("\n") if header]
if not headers and depth > 0:
headers = [""] * depth
return array, headers
def _parse_sequence_payload(data: bytes) -> np.ndarray:
seqlen = int.from_bytes(data[:_UINT32_BYTES], "little")
return np.frombuffer(data[_UINT32_BYTES:], dtype="|S1").reshape(-1, seqlen)
def _full_payload(array: np.ndarray, headers: Sequence[str]) -> bytes:
depth, seqlen = array.shape
prefix = b"".join(
(
_SERIALIZATION_VERSION.to_bytes(1, "little"),
seqlen.to_bytes(_UINT32_BYTES, "little"),
depth.to_bytes(_UINT32_BYTES, "little"),
)
)
return prefix + array.tobytes() + "\n".join(headers).encode()
def _sequence_payload(array: np.ndarray) -> bytes:
return array.shape[1].to_bytes(_UINT32_BYTES, "little") + array.tobytes()
def _random_row_indices(depth: int, count: int) -> np.ndarray:
sampled = np.random.choice(depth - 1, count - 1, replace=False) + 1
return np.sort(np.append(0, sampled))
@dataclass(frozen=True)
class FastMSA(SequentialDataclass):
"""An MSA stored as a two-dimensional NumPy byte array."""
array: np.ndarray
headers: list[str] | None = None
def __post_init__(self) -> None:
if not isinstance(self.array, np.ndarray):
raise TypeError("FastMSA array must be a NumPy array.")
if self.array.ndim != 2 or self.array.shape[0] == 0 or self.array.shape[1] == 0:
raise ValueError(
f"FastMSA array must have non-empty shape (depth, length), got {self.array.shape}."
)
if self.headers is not None and len(self.headers) != self.depth:
raise ValueError("Number of headers must match depth.")
@property
def depth(self) -> int:
return self.array.shape[0]
@property
def seqlen(self) -> int:
return self.array.shape[1]
def __len__(self) -> int:
return self.seqlen
@classmethod
def from_bytes(cls, data: bytes) -> FastMSA:
array, headers = _parse_full_payload(data)
return cls(array, headers)
@classmethod
def from_sequence_bytes(cls, data: bytes) -> FastMSA:
return cls(_parse_sequence_payload(data))
def __getitem__(
self,
indices: int | list[int] | slice | np.ndarray,
) -> FastMSA:
column_indices = [indices] if isinstance(indices, int) else indices
return dataclasses.replace(self, array=self.array[:, column_indices])
def select_sequences(
self,
indices: Sequence[int] | np.ndarray,
) -> FastMSA:
headers = None
if self.headers is not None:
headers = [self.headers[index] for index in indices]
return dataclasses.replace(
self,
array=self.array[indices],
headers=headers,
)
def select_random_sequences(self, num_seqs: int) -> FastMSA:
if num_seqs >= self.depth:
return self
return self.select_sequences(_random_row_indices(self.depth, num_seqs))
def pad_to_depth(self, depth: int) -> FastMSA:
if depth < self.depth:
raise ValueError(f"Cannot pad to depth {depth} when depth is {self.depth}")
if depth == self.depth:
return self
row_count = depth - self.depth
pad_value = ord("-") if self.array.dtype == np.uint8 else b"-"
array = np.pad(
self.array,
((0, row_count), (0, 0)),
constant_values=pad_value,
)
headers = None if self.headers is None else self.headers + [""] * row_count
return dataclasses.replace(self, array=array, headers=headers)
@classmethod
def concat(
cls,
msas: Sequence[FastMSA],
join_token: str | None = None,
allow_depth_mismatch: bool = False,
) -> FastMSA:
if not msas:
raise ValueError("Cannot concatenate an empty list of MSAs")
if join_token not in (None, ""):
raise NotImplementedError("join_token is not supported for FastMSA")
depths = [msa.depth for msa in msas]
if len(set(depths)) != 1:
if not allow_depth_mismatch:
raise ValueError("Depth mismatch in concatenating MSAs")
maximum_depth = max(depths)
msas = [msa.pad_to_depth(maximum_depth) for msa in msas]
header_columns = (
msa.headers if msa.headers is not None else [""] * msa.depth for msa in msas
)
headers = [
"|".join(str(header) for header in row) for row in zip(*header_columns, strict=False)
]
return cls(
np.concatenate([msa.array for msa in msas], axis=1),
headers,
)
@classmethod
def stack(
cls,
msas: Sequence[FastMSA],
remove_query_from_later_msas: bool = True,
) -> FastMSA:
if not msas:
raise ValueError("Cannot stack an empty list of MSAs")
arrays: list[np.ndarray] = []
headers: list[str] | None = [] if any(msa.headers is not None for msa in msas) else None
for index, msa in enumerate(msas):
start = 1 if index > 0 and remove_query_from_later_msas else 0
arrays.append(msa.array[start:])
if headers is not None:
source_headers = msa.headers or [""] * msa.depth
headers.extend(source_headers[start:])
return cls(np.concatenate(arrays, axis=0), headers)
def to_msa(self) -> MSA:
headers = self.headers
if headers is None:
headers = [f"seq{index}" for index in range(self.depth)]
entries = [
FastaEntry(header, b"".join(row).decode())
for header, row in zip(headers, self.array, strict=False)
]
return MSA(entries)
@dataclass(frozen=True)
class MSA(SequentialDataclass):
"""An ordered set of aligned protein sequences and optional A3M metadata."""
entries: list[FastaEntry]
deletions: np.ndarray | None = dataclasses.field(default=None, compare=False)
def __post_init__(self) -> None:
if not isinstance(self.entries, list):
raise TypeError("MSA entries must be a list of FastaEntry rows.")
if not self.entries:
raise ValueError("MSA requires at least one aligned sequence.")
if any(not isinstance(entry, FastaEntry) for entry in self.entries):
raise TypeError("Every MSA entry must be a FastaEntry.")
expected_length = len(self.entries[0].sequence)
if expected_length == 0:
raise ValueError("MSA sequences must be non-empty.")
for row, entry in enumerate(self.entries[1:], start=1):
if len(entry.sequence) != expected_length:
raise ValueError(
"MSA row length mismatch: "
f"row 0 has {expected_length} columns, row {row} has "
f"{len(entry.sequence)}."
)
deletions = self.deletions
if deletions is not None and not isinstance(deletions, np.ndarray):
raise TypeError("MSA deletions must be a NumPy array when provided.")
if isinstance(deletions, np.ndarray) and deletions.shape != (
len(self.entries),
expected_length,
):
raise ValueError(
"MSA deletion matrix must have shape "
f"({len(self.entries)}, {expected_length}), got {deletions.shape}."
)
@cached_property
def sequences(self) -> list[str]:
return [entry.sequence for entry in self.entries]
@cached_property
def headers(self) -> list[str]:
return [entry.header for entry in self.entries]
@property
def depth(self) -> int:
return len(self.entries)
@property
def seqlen(self) -> int:
return len(self.entries[0].sequence)
@property
def query(self) -> str:
return self.entries[0].sequence
@cached_property
def array(self) -> np.ndarray:
return np.array([list(sequence) for sequence in self.sequences], dtype="|S1")
@cached_property
def seqid(self) -> np.ndarray:
byte_array = self.array.view(np.uint8)
return (1 - cdist(byte_array[0][None], byte_array, "hamming"))[0]
def __len__(self) -> int:
return self.seqlen
def __repr__(self) -> str:
return f"MSA({self.entries[0].header}: Depth={self.depth}, Length={self.seqlen})"
@classmethod
def from_a3m(
cls,
path: PathOrBuffer,
remove_insertions: bool = True,
max_sequences: int | None = None,
) -> MSA:
entries = []
deletion_rows = []
for header, raw_sequence in islice(read_sequences(path), max_sequences):
if remove_insertions:
deletion_rows.append(a3m_deletion_counts(raw_sequence))
sequence = (
remove_insertions_from_sequence(raw_sequence) if remove_insertions else raw_sequence
)
if entries:
expected_length = len(entries[0].sequence)
if len(sequence) != expected_length:
raise ValueError(
"Sequence length mismatch. "
f"Expected: {expected_length}, Received: {len(sequence)}"
)
entries.append(FastaEntry(header, sequence))
deletions = None
if remove_insertions and deletion_rows:
deletions = np.stack(deletion_rows).astype(np.float32)
return cls(entries, deletions=deletions)
@classmethod
def from_stockholm(
cls,
path: PathOrBuffer,
remove_insertions: bool = True,
max_sequences: int | None = None,
) -> MSA:
entries = []
for record in islice(SeqIO.parse(path, "stockholm"), max_sequences):
sequence = str(record.seq)
if entries:
expected_length = len(entries[0].sequence)
if len(sequence) != expected_length:
raise ValueError(
"Sequence length mismatch. "
f"Expected: {expected_length}, Received: {len(sequence)}"
)
entries.append(FastaEntry(f"{record.id} {record.description}", sequence))
msa = cls(entries)
if remove_insertions:
msa = msa.select_positions(
[index for index, residue in enumerate(msa.query) if residue != "-"]
)
return msa
@classmethod
def from_sequences(
cls,
sequences: list[str],
remove_insertions: bool = False,
) -> MSA:
transform = (
remove_insertions_from_sequence if remove_insertions else lambda sequence: sequence
)
return cls([FastaEntry("", transform(sequence)) for sequence in sequences])
@classmethod
def from_bytes(cls, data: bytes) -> MSA:
array, headers = _parse_full_payload(data)
return cls(
[
FastaEntry(header, b"".join(row).decode())
for header, row in zip(headers, array, strict=False)
]
)
@classmethod
def from_sequence_bytes(cls, data: bytes) -> MSA:
array = _parse_sequence_payload(data)
return cls([FastaEntry("", b"".join(row).decode()) for row in array])
@classmethod
def from_state_dict(cls, dct: dict[str, Any]) -> MSA:
deletions = dct.get("deletions")
return cls(
[FastaEntry("", sequence) for sequence in dct["sequences"]],
deletions=(None if deletions is None else np.asarray(deletions, dtype=np.float32)),
)
def to_a3m(self, path: PathOrBuffer) -> None:
write_sequences(self.entries, path)
def to_fast_msa(self) -> FastMSA:
return FastMSA(self.array, self.headers)
def to_bytes(self) -> bytes:
return _full_payload(self.array, self.headers)
def to_sequence_bytes(self) -> bytes:
"""Serialize aligned sequences without their headers."""
return _sequence_payload(self.array)
def state_dict(self, json_serializable: bool = False) -> dict[str, Any]:
result: dict[str, Any] = {"sequences": self.sequences}
if self.deletions is not None:
result["deletions"] = self.deletions.tolist() if json_serializable else self.deletions
return result
def _aligned_deletions(self) -> np.ndarray | None:
if self.deletions is None:
return None
if self.deletions.shape != (self.depth, self.seqlen):
return None
return self.deletions
def _select_deletion_columns(self, indices) -> np.ndarray | None:
if self.deletions is None or self.deletions.shape[1] != self.seqlen:
return None
return self.deletions[:, indices]
def select_sequences(
self,
indices: Sequence[int] | np.ndarray,
) -> MSA:
deletions = None if self.deletions is None else self.deletions[np.asarray(indices)]
return dataclasses.replace(
self,
entries=[self.entries[index] for index in indices],
deletions=deletions,
)
def select_positions(
self,
indices: Sequence[int] | np.ndarray,
) -> MSA:
entries = [
FastaEntry(
entry.header,
"".join(entry.sequence[index] for index in indices),
)
for entry in self.entries
]
return dataclasses.replace(
self,
entries=entries,
deletions=self._select_deletion_columns(indices),
)
def __getitem__(
self,
indices: int | list[int] | slice | np.ndarray,
) -> MSA:
column_indices = [indices] if isinstance(indices, int) else indices
entries = [
FastaEntry(
entry.header,
slice_any_object(entry.sequence, column_indices),
)
for entry in self.entries
]
return dataclasses.replace(
self,
entries=entries,
deletions=self._select_deletion_columns(column_indices),
)
def greedy_select(self, num_seqs: int, mode: str = "max") -> MSA:
if mode not in ("max", "min"):
raise ValueError(f"Unsupported MSA selection mode: {mode!r}.")
if self.depth <= num_seqs:
return self
return self.select_sequences(greedy_select_indices(self.array, num_seqs, mode))
def hhfilter(
self,
seqid: int = 90,
diff: int = 0,
cov: int = 0,
qid: int = 0,
qsc: float = -20.0,
binary: str = "hhfilter",
) -> MSA:
indices = hhfilter(
self.sequences,
seqid=seqid,
diff=diff,
cov=cov,
qid=qid,
qsc=qsc,
binary=binary,
)
return self.select_sequences(indices)
def select_random_sequences(self, num_seqs: int) -> MSA:
if num_seqs >= self.depth:
return self
return self.select_sequences(_random_row_indices(self.depth, num_seqs))
def select_diverse_sequences(self, num_seqs: int) -> MSA:
if num_seqs >= self.depth:
return self
filtered = self.hhfilter(diff=num_seqs)
if num_seqs < filtered.depth:
filtered = filtered.select_random_sequences(num_seqs)
return filtered
def pad_to_depth(self, depth: int) -> MSA:
if depth < self.depth:
raise ValueError(f"Cannot pad to depth {depth} when depth is {self.depth}")
if depth == self.depth:
return self
count = depth - self.depth
extra = [FastaEntry("", "-" * self.seqlen) for _ in range(count)]
deletions = self._aligned_deletions()
if deletions is not None:
zero_rows = np.zeros((count, self.seqlen), dtype=deletions.dtype)
deletions = np.concatenate((deletions, zero_rows), axis=0)
return dataclasses.replace(
self,
entries=self.entries + extra,
deletions=deletions,
)
@classmethod
def stack(
cls,
msas: Sequence[MSA],
remove_query_from_later_msas: bool = True,
) -> MSA:
entries = []
deletion_arrays = []
for index, msa in enumerate(msas):
start = 1 if index > 0 and remove_query_from_later_msas else 0
entries.extend(msa.entries[start:])
aligned = msa._aligned_deletions()
if aligned is not None:
deletion_arrays.append(aligned[start:])
deletions = None
if (
len(deletion_arrays) == len(msas)
and len({array.shape[1] for array in deletion_arrays}) == 1
):
deletions = np.concatenate(deletion_arrays, axis=0)
return cls(entries=entries, deletions=deletions)
@classmethod
def concat(
cls,
msas: Sequence[MSA],
join_token: str | None = "|",
allow_depth_mismatch: bool = False,
) -> MSA:
if not msas:
raise ValueError("Cannot concatenate an empty list of MSAs")
depths = [msa.depth for msa in msas]
if len(set(depths)) != 1:
if not allow_depth_mismatch:
raise ValueError("Depth mismatch in concatenating MSAs")
maximum_depth = max(depths)
msas = [msa.pad_to_depth(maximum_depth) for msa in msas]
headers = [
"|".join(str(header) for header in row)
for row in zip(*(msa.headers for msa in msas), strict=False)
]
separator = "" if join_token is None else join_token
sequences = [
separator.join(row) for row in zip(*(msa.sequences for msa in msas), strict=False)
]
deletions = None
if separator == "":
arrays = [msa._aligned_deletions() for msa in msas]
if all(array is not None for array in arrays):
deletions = np.concatenate(arrays, axis=1) # type: ignore[arg-type]
return cls(
[
FastaEntry(header, sequence)
for header, sequence in zip(headers, sequences, strict=False)
],
deletions=deletions,
)
|