Novel method to bias ProteinMPNN for antibody design, without modifying model weights.
Logits from protein-general ProteinMPNN and antibody-specific AbLANG are added and softmaxed. Addition of AbLANG is supposed to push the model into the antibody-acceptable space.
On in-silico experiments ProteinMPNN+AbLang outperformed ProteinMPNN alone and rivalled antibody-specific AbMPNN.
Authors designed 96 variants of Trastuzumab CDR-H3 using ProteinMPNN, AbLang and ProteinMPNN+AbLang each. AbLANG and ProteinMPNN produced 1 and 3 successful variants respecitively (both out of 96) whereas their combination produced 36 successful variants.
None of the variants were better variants than WT Trastuzumab.
PSBench is a large benchmark dataset (>1M models) for training and evaluating model accuracy estimation (EMA) methods for protein complex structures, using data from CASP15 & CASP16.
Models were generated by AlphaFold2-Multimer and AlphaFold3 under blind prediction conditions and annotated with 10 detailed global, local, and interface quality scores.
The dataset enables development of advanced EMA methods (e.g. GATE), which showed top performance in blind CASP16 assessments.
The model can do autoregressive generation N-to-C, C-to-N, and also supports span infilling.
The architecture is a Transformer with a Sparse Mixture of Experts (MoE), activating about 27% of parameters per forward pass to improve computational efficiency.
They studied how sampling affects training by trying different family-level weighting schemes. Uniform sampling across families (where small and large families have equal chance) gave better diversity and generalization, while unmodified sampling (letting big families dominate) performed worst.
They validated the models by showing that generated proteins express well in wet lab experiments (split-GFP assays, spanning both highly novel and moderately novel sequence spaces).
They used a large thermostability dataset to align model predictions to stability. This alignment is not standard fine-tuning — instead, preference optimization was applied, teaching the model to prefer sequences predicted to have higher stability. Upon experimental validation, aligned models indeed produced proteins with higher expression and stability.
Improvement upon earlier RFDiffusion, enhancing stability and accuracy in designing enzyme active sites.
Catalytic sites can now be specified at the atomic level instead of the residue backbone level used previously. This eliminates the need to explicitly enumerate side-chain rotamers.
Training uses flow matching, a technique that simplifies and stabilizes the diffusion training process.
Benchmarked on a set of 41 diverse enzyme active sites; RFdiffusion2 succeeded in all 41 cases, significantly outperforming the earlier RFDiffusion, which succeeded in only 16.
Boltz-1 is an open-source reproduction of AlphaFold3, which uses a diffusion module to co-fold molecular structures (proteins, ligands, etc.).
For design purposes, BoltzDesign1 sidesteps the full structure generation step and instead uses only the Pairformer (which outputs a distogram — a probabilistic representation of all pairwise residue distances). This allows broader exploration of sequence space, as it optimizes over the distribution of possible structures rather than committing to a single conformation.
Given a target (such as a small molecule or protein), they weakly initialize a binder sequence using random logits. This sequence is then iteratively refined by backpropagating loss through the Pairformer (and optionally through the Confidence module) to increase the predicted quality of the binder–target interaction.
A full 3D structure can be generated at the end using the Boltz-1 structure module, but this is not part of the optimization loop.
They benchmarked their method in silico on small molecule targets and a set of protein–protein interactions from the BindCraft benchmark, comparing performance to RfDiffusion All-Atom.
BindCraft is an easy-to-use pipeline for computational protein binder design.
It employs AlphaFold2-Multimer to hallucinate binders via backpropagation.
Given a target structure and binder parameters (e.g., sequence length), the binder sequence is initialized with random logits and iteratively optimized via gradient descent through the AF2-Multimer network.
After binder hallucination, the sequence and surface residues are further optimized using MPNNsol, and AF2-Monomer is used to repredict and filter high-confidence designs.
Binder designs were validated experimentally through in vitro assays, X-ray crystallography, and cryo-EM.
Reported success rates ranged from 25% to 100%, with most binders in the nanomolar affinity range, a few in the micromolar range, and backbone RMSDs of ~1.7 Å to 3.1 Å between design models and solved structures.