Reproduce “NeuralPLexer3: Accurate Biomolecular Complex Structure Prediction with Flow Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhuoran Qiao, Feizhi Ding, Thomas Dresselhaus, Mia A. Rosenfeld, Xiaotian Han, Owen Howell, Aniketh Iyengar, Stephen Opalenski and 5 more
Iambic Therapeutics developed NeuralPLexer3 (NP3), a physics-inspired flow-based generative model for biomolecular complex structure prediction, achieving state-of-the-art accuracy, including 78.4% physically valid predictions on PoseBusters, while operating 15x faster than AlphaFold3 at approximately 30 seconds per prediction on a single L40S GPU.
Key insights
- Integrating physics-motivated priors and optimal transport principles into flow-based generative models significantly improves the physical validity and stereochemical accuracy of predicted biomolecular structures.
- Strategic hardware-aware optimizations, such as Flash-TriangularAttention and efficient flow samplers, can achieve substantial speedups (15x faster than AlphaFold3) without sacrificing prediction accuracy, making AI models practical for high-throughput drug discovery.
- Newly introduced benchmarks, NPBench and ConfBench, are crucial for rigorously evaluating model performance on diverse biomolecular interactions and ligand-induced conformational changes, providing more relevant metrics for drug discovery applications.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.