Genomics and RNA

Edit and crop an alternate-allele DNA window

MediumRaw PyTorchAbout 20 min1 public example plus private tests

Free account with email, Google or GitHub. Your code runs in an isolated PyTorch sandbox.

Your task

Replace a reference DNA interval with an alternate allele and crop the requested fixed-width window. Return its sequence, validity, and source-position maps.

Function signature

allele_window_edit(
    reference: torch.Tensor,
    alternate: torch.Tensor,
    reference_valid: torch.Tensor,
    alternate_valid: torch.Tensor,
    options: dict,
) -> dict[str, torch.Tensor]

Worked example

Insertion edit

In the first row, reference ATCATT has its C at zero-based index 2 replaced by ATG. The full edited sequence is ATATGATT, of length 6 + 3 − 1 = 8. Output slots 2–4 come from alternate indices 0–2; surviving reference slots keep indices 0, 1, 3, 4, 5.

import torch

reference = torch.tensor([[[1.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 1.0], [0.0, 1.0, 0.0, 0.0], [1.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 1.0], [0.0, 0.0, 0.0, 1.0]], [[0.0, 0.0, 0.0, 1.0], [0.0, 0.0, 0.0, 1.0], [0.0, 1.0, 0.0, 0.0], [0.0, 0.0, 0.0, 1.0], [0.0, 1.0, 0.0, 0.0], [0.0, 0.0, 1.0, 0.0]]], dtype=torch.float64, requires_grad=True)
alternate = torch.tensor([[[1.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 1.0], [0.0, 0.0, 1.0, 0.0]], [[1.0, 0.0, 0.0, 0.0], [1.0, 0.0, 0.0, 0.0], [1.0, 0.0, 0.0, 0.0]]], dtype=torch.float64, requires_grad=True)
reference_valid = torch.full((2, 6), True, dtype=torch.bool)
alternate_valid = torch.tensor([[True, True, True], [True, True, True]], dtype=torch.bool)
options = {
    'anchor': 2,
    'ref_length': 1,
}

result = allele_window_edit(reference=reference, alternate=alternate, reference_valid=reference_valid, alternate_valid=alternate_valid, options=options)

Expected return

{
    'sequence': torch.tensor([[[1.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 1.0], [1.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 1.0], [0.0, 0.0, 1.0, 0.0], [1.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 1.0], [0.0, 0.0, 0.0, 1.0]], [[0.0, 0.0, 0.0, 1.0], [0.0, 0.0, 0.0, 1.0], [1.0, 0.0, 0.0, 0.0], [1.0, 0.0, 0.0, 0.0], [1.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 1.0], [0.0, 1.0, 0.0, 0.0], [0.0, 0.0, 1.0, 0.0]]], dtype=torch.float64),
    'valid_mask': torch.full((2, 8), True, dtype=torch.bool),
    'reference_index': torch.tensor([0, 1, -1, -1, -1, 3, 4, 5], dtype=torch.int64),
    'alternate_index': torch.tensor([-1, -1, 0, 1, 2, -1, -1, -1], dtype=torch.int64),
}

Parameters

Parameter	Shape and meaning
`reference`torch.Tensor	$[B, L, 4]$ `float` $[B,L,4]$ one-hot or soft DNA payloads in A, $C$ , $G$ , $T$ order
`alternate`torch.Tensor	$[B, A, 4]$ `float` $[B,A,4]$ complete replacement allele, including retained anchor when applicable
`reference_valid`torch.Tensor	$[B, L]$ `bool` $[B,L]$ , source-window support
`alternate_valid`torch.Tensor	$[B, A]$ `bool` $[B,A]$ , allele-payload support
`options`dict	Scalar settings. Omitted keys use the defaults below.

Return

dict[str, torch.Tensor]

Return exactly the named fields applicable to this operation.

Required behavior

reference $[B,L,4]$ and alternate $[B,A,4]$ use channel order $[A,C,G,T]$ . Config $\mathrm{anchor}=P$ is a zero-based index into the supplied reference window; $\mathrm{ref\_length}=R$ is the positive number of replaced bases, including a retained anchor when using normalized VCF-style alleles.
Require $A\ge 1$ , $0\le P$ , $P+R\le L$ . Form $\mathrm{edited}=\mathrm{reference}[:P] + \mathrm{alternate} + \mathrm{reference}[P+R:]$ , of length $\mathrm{Le}=L+A-R$ , after sanitizing missing payloads via reference_valid/alternate_valid.
Config window_start (default $0$ ) indexes this EDITED sequence, and window_length (defaultLe) may crop or pad the requested window; negative start or coordinates beyond Le produce zero DNA and valid_mask=False. A biologically unknown but valid base can be an all-zero vector with valid_mask=True.
Return sequence $[B,W,4]$ , valid_mask $[B,W]$ , and payload provenance arrays: reference_index $[W]$ gives a retained original-input coordinate, alternate_index $[W]$ gives a supplied alternate-allele coordinate; $\mathrm{use} -1$ in the other array and for padding.
The supplied alternate's retained anchor is still alternate payload, not falsely labeled reference provenance. $W=0$ is permitted. This primitive edits tensor payloads, not strings or a reference-genome database.
All entries in the batch use the same anchor, ref_length, window_start and window_length; only sequence payloads and masks differ.

Why this matters

An insertion or deletion changes sequence length, so payloads and their provenance must move through the same edit and crop.

Starter code

You write the body of this function in the BioTorch workspace. Your code runs in an isolated PyTorch sandbox; Run checks the public examples and Submit grades the full test suite, including gradients.

import torch


def allele_window_edit(
    reference: torch.Tensor,  # (B, L, 4)
    alternate: torch.Tensor,  # (B, A, 4)
    reference_valid: torch.Tensor,  # (B, L)
    alternate_valid: torch.Tensor,  # (B, A)
    options: dict,
) -> dict[str, torch.Tensor]:
    """Return dict[str, torch.Tensor]:
    
    sequence: (B, W, 4).
    valid_mask: (B, W).
    reference_index: (W,).
    alternate_index: (W,)."""
    # Implement the operation. Inputs and test setup are supplied by BioTorch.
    raise NotImplementedError("Implement allele_window_edit")

Ready to implement it?

3 progressive hints, a solution walkthrough, and grading that checks values, shapes and gradients are waiting in the workspace.

Write and run your solution Sign in

Free account with email, Google or GitHub. Your code runs in an isolated PyTorch sandbox.

Sources and further reading

A primitive exercise, not an end-to-end reproduction of the named models. No affiliation or endorsement is implied.