#5 · Structural bio & protein design
MD ensembles and replica-exchange sampling map folding and binding pathways
The common thread is treating molecular recognition and folding as sampled ensembles, pathways, and landscapes rather than single endpoint structures. Markov-model and VAE pathway clustering of MD simulations maps de novo folding of 20 lasso peptides onto uphill landscapes with entropy-driven steps, while two-dimensional replica-exchange MD (gREST/REUS) resolves encounter regions and intermediates between Abltide's initial approach and its bound pose on Abl kinase. What is new is the emphasis on intermediate-rich landscapes over endpoints. Other items apply related tools with different emphases: MD with alchemical free energy calculations for ochratoxin A aptamer selectivity, RFdiffusion auxiliary potentials for nucleic acid binders, and an AlphaFold3-MD framework that redesigns a fragile α-helical domain into a four-helix bundle.
Preprints
Evaluating and improving AI agents for sequencing protocol understanding
LibStructBench, a document-grounded benchmark testing whether AI agents can exactly reconstruct sequencing-library structures and molecular workflows from…
Do Language Models Know What Compelling Scientific Evidence Looks Like?
Introduces EDB-Bio, a 96k-episode benchmark from drug transcriptomics, CRISPR screens and GWAS testing whether LMs weigh biological evidence; post-training…
Harness engineering enables a multi-agent LLM system to generate cross-domain hypotheses beyond single-investigator reach
Prompting harness, not model tier, drives quality of a biology multi-agent LLM hypothesis generator; introduces ABCD-Stack.
Topic history
| Date | Rank |
|---|---|
| #5 | |
| #5 | |
| #4 | |
| #1 | |
| #1 | |
| #3 | |
| #2 | |
| #1 | |
| #1 | |
| #2 | |
| #4 | |
| #4 | |
| #5 | |
| #13 | |
| #6 | |
| #4 | |
| #3 | |
| #7 | |
| #6 |