PredIDR3: A new output-encoding scheme and abundant negative source provide more information for deep learning-based protein intrinsic disorder prediction [new]

Show Post

Retrieval of binding sites across the AlphaFold human proteome using protein language model representations [new] ...by using frozen ESM-C 600M embeddings of cavity-lining residues for exhaustive MaxSim retrieval across the proteome.

Show Post

Because corridor construction is decoupled from DP scoring, it outputs explicit [Li, Ri] bounds to act as a front-end pruner for fast C/SIMD/GPU banded backends reduing both the DP search space and memory footprint. @albertvilella.bsky.social @darwinianascent.bsky.social

Show Post

AB-NW uses a coarse PLM embedding landscape to prune the DP search space upfront. This yields 55–91% DP cell savings while retaining ≥99% exact unconstrained alignment accuracy.

Show Post

However, AB-NW targets the protein twilight zone (<30% identity), large structural indels, and MSA guide trees where exact seeds fail, static bands recover only 20–50% of core blocks, and existing PLM aligners evaluate full dense O(NM) matrices.

Show Post