#5 · Genomics
Computational models link germinal center B cell states to antibody affinity maturation
The dominant thread is computational modeling of antibody affinity maturation, particularly at the germinal center level. Ancestra provides a lineage-explicit simulator of BCR heavy-chain affinity maturation that yields sequence datasets with ground-truth lineage trees, enabling benchmarking of repertoire and lineage inference methods. A germinal center model reports that phased fragility and stability of non-genetic B cell states accelerate the genetic evolution of antibodies. A separate eLife paper proposes a biologically informed training paradigm for protein language models that predicts affinity maturation trajectories. What is new across these items is the movement from purely sequence-level or lineage-level descriptions toward models that carry explicit lineage truth, non-genetic state dynamics, and biological constraints on learning. The items are heterogeneous in scope, and the link between benchmarking and prediction remains a proposed rather than demonstrated connection.
Published papers
SAEKP: A Site Attention Enzyme Kinetics Prediction Method to Facilitate Enzyme Engineering.
SAEKP predicts enzyme kinetic parameters (kcat, Km, Ki) using multimodal enzyme and substrate representations plus a site attention module that reweights…
Predicting Protein Thermostability with Equivariant Graph Neural Networks and Contrastive Learning.
ECL-Net combines protein language model embeddings, residue interaction graphs, EGNN, and contrastive learning to predict thermostability changes from…
AdventML: Advanced Enzyme Temperature Prediction with Transformer-Based Embeddings and Resampling Strategies.
AdventML, an ML framework predicting enzyme optimal temperature (Topt) from transformer protein embeddings, with resampling for extremes; R2 ≈ 0.64,…
ESMRank: learning-to-rank prediction of missense variant effects from deep mutational scanning data.
ESMRank: a learning-to-rank missense variant predictor using overlap-aware rank fusion of ~1100 deep mutational scans, outperforming baselines on ProteinGym…
Domain-Aware Enzyme Classification via Multiple Instance Learning.
MIL-BC models proteins as bags of automatically detected domains for EC number prediction, improving F1 by 0.03-0.17 over a sequence baseline without…
Preprints
MPFIL: Multi-Level Protein Feature Integration for Predicting Mutation-Induced Protein-Protein Binding Affinity Changes
MPFIL integrates structural and sequence features of wild-type and mutant protein complexes to predict mutation-induced changes in protein-protein binding…
AltereGO: Prediction of Protein Function by Multimodal Fusion of Protein Language Models.
AltereGO fuses score-level predictions from ESM2-3B, ProtT5, Ankh and ProstT5-3Di protein representations to predict Gene Ontology molecular function,…
High-throughput mapping of kinase-substrate interactions with protein language models and phospho-PCA
KINBERT, a protein language model paired with phospho-PCA pooled screening and Bayesian hierarchical modeling, maps kinase-substrate interactions, including…
Topic history
| Date | Rank |
|---|---|
| #5 | |
| #5 | |
| #1 | |
| #1 | |
| #5 | |
| #3 | |
| #4 | |
| #4 | |
| #3 | |
| #2 | |
| #2 | |
| #5 | |
| #15 | |
| #11 | |
| #11 |