Structure confidence heads

Interface pLDDT Summary

EasyRaw PyTorchAbout 10 min2 public examples plus private tests

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

Your task

Summarize supplied atom-level pLDDT scores near cross-chain interfaces, with the specified extra weight for ligand atoms.

Function signature

interface_plddt_summary(
    plddt: torch.Tensor,
    activations: dict,
    masks: dict,
    structure: dict,
    count_eps: float = 1e-08,
    interface_cutoff: float = 8.0,
) -> torch.Tensor

Worked example

Give ligand atoms twice the weight

The tokens are $4$ units apart across chains, inside cutoff $8$ . The protein atom with pLDDT $60$ has contact weight $1$ ; the ligand atom with pLDDT $90$ has weight $2$ . The weighted mean is $(60+2\times90)/(1+2+\epsilon)$ , approximately $80$ with the default small denominator epsilon.

import torch

plddt = torch.tensor([[[60.0], [90.0]]], dtype=torch.float64, requires_grad=True)
activations = {
    'representative_distances': torch.tensor([[[0.0, 4.0], [4.0, 0.0]]], dtype=torch.float64),
}
masks = {
    'token_mask': torch.tensor([[True, True]], dtype=torch.bool),
    'atom_mask': torch.tensor([[[True], [True]]], dtype=torch.bool),
    'ligand_mask': torch.tensor([[False, True]], dtype=torch.bool),
}
structure = {
    'chain_ids': torch.tensor([[0, 1]], dtype=torch.int64),
}
count_eps = 1e-08
interface_cutoff = 8.0

result = interface_plddt_summary(plddt=plddt, activations=activations, masks=masks, structure=structure, count_eps=count_eps, interface_cutoff=interface_cutoff)

Expected return

torch.tensor([79.99999973333334], dtype=torch.float64)

Parameters

Parameter	Shape and meaning
`plddt`torch.Tensor	$[B,N,A]$ Supplied atom-level pLDDT scores. B is batch size, N tokens, and A atom slots.
`activations`dict	Supplied intermediate tensors.
`masks`dict	Supplied masks. Each field below defines what True includes or permits.
`structure`dict	Fields supplied together in this dictionary.
`count_eps`floatDefault `1e-08`	Positive offset in the declared TM/confidence weighted-count denominators, added after summing the relevant support.
`interface_cutoff`floatDefault `8.0`	Distance threshold for selecting interface neighborhoods.

Return

torch.Tensor · $[B]$

Return Weighted interface confidence mean per structure.

Required behavior

A token has an interface contact if any valid token in a different chain has representative_distance < interface_cutoff (default 8.0). The cutoff is strict; distances are supplied, not recomputed from atoms.
A ligand token has weight 2 whether or not it has a contact. Every other token has weight 1 if it has a contact and 0 otherwise. Apply that weight to every valid atom of the token; invalid token/atom slots have weight 0.
Return sum(plddt × atom_weight) divided by sum(atom_weight) + count_eps (default 1e-8), separately per structure. No included atom yields zero. Do not rescale the supplied pLDDT scores.
Use CPU float32/float64 tensors with one shared floating dtype/device. Preserve strided layouts and do not mutate supplied tensors.
Use raw tensor operations, not a ready-made implementation of the requested primitive. Supplied parameters and random variates must not be reinitialized.

Why this matters

Interface confidence summarizes atom-level confidence near contacts between chains. This teaching rule gives ligand atoms twice the weight of other interface atoms.

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 interface_plddt_summary(
    plddt: torch.Tensor,  # (B,N,A)
    activations: dict,
    masks: dict,
    structure: dict,
    count_eps: float = 1e-08,
    interface_cutoff: float = 8.0,
) -> torch.Tensor:
    """Return a tensor with shape (B,)."""
    # Implement the operation. Inputs and test setup are supplied by BioTorch.
    raise NotImplementedError("Implement interface_plddt_summary")

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.

Related exercises

Sources and further reading

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