Researchers introduce EnsembleEGNN, a molecular ensemble foundation model that encodes conformational ensembles of cyclic peptides using Equivariant Graph Neural Networks with Set Attention Block pooling. The model is pretrained on the CREMP cyclic peptide ensemble dataset via multi-task self-supervised objectives including masked token recovery and coordinate reconstruction. On the CREMP-CycPeptMPDB benchmark, the pretrained model achieves R²=0.477 versus complete failure (R²=0.005) when trained from scratch, and a hybrid model combining EnsembleEGNN with a BERT sequence encoder reaches R²=0.538, demonstrating that thermodynamically-informed conformational ensemble embeddings meaningfully improve property prediction.
EvoStruct addresses vocabulary collapse in GNN-based antibody CDR design by combining a frozen protein language model with an E(3)-equivariant GNN through a cross-attention adapter. The method introduces progressive PLM unfreezing and R-Drop consistency regularization to recover functionally important amino acid diversity. On CHIMERA-Bench, EvoStruct improves sequence recovery by 16%, reduces perplexity by 43%, and achieves 2.3x greater amino acid diversity compared to the best GNN baselines.
PLAID is a generative model that simultaneously produces protein 1D sequences and 3D all-atom structures by learning a diffusion model over the latent space of ESMFold, a protein folding model. It requires only sequence data for training—leveraging databases 2-4 orders of magnitude larger than structure databases—and decodes structure at inference via frozen folding model weights. The approach supports compositional prompting by function and organism, addressing practical drug-design constraints like humanization and solubility. A companion compression model, CHEAP, addresses the high-dimensionality of transformer latent spaces to make the diffusion training tractable.
COGENT is a new architecture combining graph neural networks with Neural Ordinary Differential Equations for continuous-time physical forecasting on irregular geospatial meshes. The model encodes historical system states and forcings into latent dynamics that can be queried at arbitrary future times, avoiding the error accumulation of autoregressive rollout. Evaluated on ice-sheet simulations from the Ice-sheet and Sea-level System Model, COGENT shows improved long-range stability over autoregressive graph baselines. The work introduces training stabilization strategies including rollout-horizon sampling and progressive scheduling.
The paper introduces MSN (Magnetic Structure Network), an E(3) equivariant graph neural network that predicts collinear and non-collinear magnetic structures directly from atomic crystal coordinates. Trained on experimentally determined structures from the MAGNDATA database, it uses a novel Primitive Modulated Structure Representation (PMSR) to handle both commensurate and incommensurate magnetic orders in a unified framework without symmetry assumptions. The model achieves near-experimental accuracy across diverse magnetic structure types, offering a scalable alternative to costly experiments and computationally demanding first-principles methods for magnetic materials discovery.
Researchers introduce PromptGNN-sim, a bidirectional structure-semantic fusion framework that jointly trains a Graph Attention Network and an LLM for text-attributed graph learning. The system uses GAT-based neighborhood selection to generate structure-aware prompts for the LLM, with cross-modal contrastive learning and cross-attention aligning both components during training. Evaluated on six datasets including Cora, Pubmed, and WikiCS, it outperforms classical GNNs, standalone LLMs, and prior GNN-LLM fusion methods on cross-task transfer, cross-dataset generalization, and sparse perturbation settings.
CARVE (Content-Aware Recurrent with Value Efficiency) is a new linear attention architecture that addresses three coupled defects in the GDN-2 delta-rule architecture by restricting erasure to the key axis rather than the value axis. This design choice is proven necessary and sufficient to enable the WY-form triangular chunk solver, enabling competitive training throughput with Transformers. At 1.3B parameters trained on 100B tokens, CARVE achieves lower perplexity than GDN-2, leads recurrent baselines on nine commonsense reasoning benchmarks, and sets state-of-the-art on RULER retrieval probes, while using 13% less peak memory and 19% fewer parameters at 0.4% throughput overhead.
Researchers introduce SciReasoner, a multimodal scientific foundation model that discretizes molecular coordinates, topologies, and periodic connectivities into a unified structure-aware vocabulary for reasoning over proteins, small molecules, and inorganic crystals. The model achieves state-of-the-art performance on 67 of 86 benchmarks, with notable gains in Gene Ontology prediction (F_max 0.42→0.55), retrosynthesis accuracy (0.63→0.72), and materials phase separation. Expert evaluators rated its reasoning traces as preferred or comparable to a frontier LLM in 98% of cases, suggesting meaningful interpretability alongside accuracy gains.
Biohub and EvolutionaryScale released ESMFold2, a 6.2-billion-parameter open-weights model for predicting the 3D shapes of proteins, DNA, RNA, and small molecules by treating molecular sequences as language. Unlike AlphaFold 3, ESMFold2 can operate without multiple sequence alignments (MSAs) by using a transformer-based embedding model (ESMC) trained on 2.8 billion sequences, outperforming Chai-1 in MSA-free settings and matching AlphaFold 3 when MSAs are provided. The model weights are freely available on HuggingFace and via API through Biohub, making frontier-level structural biology accessible without proprietary infrastructure. The release is significant for drug discovery involving novel or synthetic molecules where MSA databases may be sparse.