42votes

How to avoid GPU OOM when predicting large multimer complexes (>2500 residues) in AlphaFold 3?

2 hours ago
Asked by Elena_V (Structural Bioinformatician @ Oxford)

We are attempting to fold a 4-chain ternary complex (~2600 residues total) using the local AF3 pipeline on a single NVIDIA A100 80GB SXM. The pairformer attention matrix explodes during the diffusion head evaluation, throwing CUDA out of memory. Has anyone found working chunking or CPU-offloading flags?

#AlphaFold3#PyTorch#GPU-Memory#HPC-Slurm

Answers (2)

Dr. John_M · Core ML Engineer @ BioAtlas
✓ Accepted Solution

You can enable chunked pairformer cross-attention and gradient checkpointing inside the model runner config. In your run script or Python wrapper, set `pairformer_chunk_size=128` and offload the MSA sequence representations after the initial embedding pass:

# Modify runner config in model_eval.py:
model_config = AlphaFold3Config(
    pairformer_chunk_size=128,
    gradient_checkpointing=True,
    offload_msa_to_cpu=True,
    diffusion_batch_size=1
)
predictions = model.predict(multimer_features)
BioHPC_Admin · Cluster Admin @ Karolinska

If you have access to multi-GPU nodes, AF3 supports sequence parallelism across 2x A100/H100 cards via PyTorch FSDP or DeepSpeed ZeRO-3.

29votes

ESM-3 zero-shot fitness prediction vs deep mutational scanning (DMS) validation

1 day ago
Asked by Marcus_Bio (Postdoc @ Stanford Bio-X)

How does ESM-3 98B perform on predicting single-point mutant fitness compared to ESM-2 3B and Tranception on the ProteinGym benchmark? Are the geometric coordinates necessary for top correlation?

#ESM3#DMS#Protein-Design#Benchmarks
19votes

DiffDock ligand poses clash with covalent cysteine warheads — best remedy?

3 days ago
Asked by DrugHunter_42 (Medicinal Chemist @ Vertex)

We are screening covalent acrylamide inhibitors against a target with Cys145. DiffDock produces binding poses where the warhead carbon is positioned >4.5Å from the thiol sulfur, or clashes into adjacent backbone carbonyls. Is there a constrained diffusion mode?

#DiffDock#SmallMolecules#Docking#RDKit