TriFit: Trimodal Fusion with Protein Dynamics for Mutation Fitness Prediction

Seungik Cho

cs.LGApr 13, 2026

TriFit: Trimodal Fusion with Protein Dynamics for Mutation Fitness Prediction

Authors: Seungik Cho

Organizations: Department of Physics and Astronomy, Rice University, Texas, USA.

Abstract

Predicting the functional impact of single amino acid substitutions (SAVs) is central to understanding genetic disease and engineering therapeutic proteins. While protein language models and structure-based methods have achieved strong performance on this task, they systematically neglect protein dynamics; residue flexibility, correlated motions, and allosteric coupling are well-established determinants of mutational tolerance in structural biology, yet have not been incorporated into supervised variant effect predictors. We present TriFit, a multimodal framework that integrates sequence, structure, and protein dynamics through a four-expert Mixture-of-Experts (MoE) fusion module with trimodal cross-modal contrastive learning. Sequence embeddings are extracted via masked marginal scoring with ESM-2 (650M); structural embeddings from AlphaFold2-predicted C-alpha geometries; and dynamics embeddings from Gaussian Network Model (GNM) B-factors, mode shapes, and residue-residue cross-correlations. The MoE router adaptively weights modality combinations conditioned on the input, enabling protein-specific fusion without fixed modality assumptions. On the ProteinGym substitution benchmark (217 DMS assays, 696k SAVs), TriFit achieves AUROC 0.897 +/- 0.0002, outperforming all supervised baselines including Kermut (0.864) and ProteinNPT (0.844), and the best zero-shot model ESM3 (0.769). Ablation studies confirm that dynamics provides the largest marginal contribution over pairwise modality combinations, and TriFit achieves well-calibrated probabilistic outputs (ECE = 0.044) without post-hoc correction.

Figures & tables

Figure 1: TriFit architecture. Sequence, structure, and dynamics encoders (frozen) extract modality-specific embeddings. A learned projection maps each to a shared 512-dim space. The four-expert MoE router adaptively combines modality pairs (E1: Seq+Struct, E2: Seq+Dyn, E3: Struct+Dyn, E4: Trimodal) via soft gating. Cross-modal contrastive loss aligns all three modality pairs during training. The weighted-fused representation $\mathbf{f}$ is passed to a binary classifier predicting functional vs. damaging variants.

Model	AUROC	AUPRC	ACC	mF1	mAR	mAP
OHE (no aug.)	0.6387 ±.0434	0.6990 ±.1308	0.5901 ±.0324	0.5760 ±.0480	0.6034 ±.0327	0.5901 ±.0324
OHE + ESM-1v	0.7581 ±.0959	0.7838 ±.1429	0.6692 ±.0695	0.6569 ±.0818	0.6922 ±.0725	0.6691 ±.0695
OHE + MSA-T	0.7644 ±.0828	0.7968 ±.1322	0.6734 ±.0625	0.6612 ±.0763	0.6967 ±.0640	0.6734 ±.0625
OHE + TranceptEVE	0.7744 ±.0781	0.8051 ±.1301	0.6799 ±.0589	0.6680 ±.0723	0.7054 ±.0600	0.6798 ±.0589
Emb + ESM-1v	0.8062 ±.1009	0.8170 ±.1432	0.7007 ±.0765	0.6892 ±.0888	0.7281 ±.0760	0.7006 ±.0765
Emb + MSA-T	0.8334 ±.0873	0.8478 ±.1258	0.7217 ±.0735	0.7105 ±.0875	0.7506 ±.0690	0.7216 ±.0735

Table 1: Comparison with supervised baselines on ProteinGym (217 proteins). Best results in bold , second best underlined . Mean

\pm

std over 3 seeds.

Single modality
Configuration	AUROC	AUPRC	ACC	mF1	mAR	mAP
Seq only	0.8743 ±.0006	0.9068 ±.0006	0.8039 ±.0004	0.7995 ±.0002	0.7988 ±.0003	0.8003 ±.0006
Struct only	0.8376 ±.0003	0.8638 ±.0006	0.7633 ±.0006	0.7561 ±.0008	0.7541 ±.0010	0.7597 ±.0005
Dyn only	0.7838 ±.0006	0.8301 ±.0004	0.7116 ±.0006	0.7026 ±.0011	0.7011 ±.0013	0.7059 ±.0007
Pairwise modality
Seq + Struct	0.8759 ±.0001	0.9083 ±.0002	0.8071 ±.0008	0.8024 ±.0007	0.8013 ±.0012	0.8040 ±.0014

Table 2: Ablation study on ProteinGym test set (mean

\pm

std, 3 seeds).

Figure 2: Representation analysis. Left: UMAP of projected modality embeddings (sequence: orange, structure: blue, dynamics: green). Right: LDA projection of MoE fused embeddings showing distributional shift between damaging (red) and functional (blue) variants.

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Model	Type	AUROC
Site Independent	Zero-shot	0.6967
ProteinMPNN	Zero-shot	0.6532
EVmutation	Zero-shot	0.7165
ESM1b	Zero-shot	0.7212
ESM1v	Zero-shot	0.7166
ESM1v (ensemble)	Zero-shot	0.7312

Appendix

Table 3: Zero-shot baseline comparison on ProteinGym (AUROC only; max(AUROC, 1-AUROC) reported for zero-shot models to account for score direction). TriFit is supervised.

Figure 3: MoE expert utilization across 217 test proteins (hierarchical clustering by router weight pattern). Color intensity indicates mean router weight assigned to each expert. Two major clusters emerge: proteins preferring the Trimodal expert (E4) and proteins preferring the Struct+Dyn expert (E3). The Seq+Dyn expert (E2) is consistently underweighted, while Seq+Struct (E1) shows intermediate utilization.

Figure 4: Prediction calibration analysis on the ProteinGym test set (139,480 variants). Left: Reliability diagram showing close alignment between predicted probabilities and empirical positive rates (ECE $=0.044$ ), achieved without post-hoc calibration. Center: Confidence distribution $\max(p,1{-}p)$ by true label, showing similar confidence profiles for both classes. Right: Confidence vs. accuracy across 10 equal-width bins; dot size proportional to bin count.

Figure 5: Per-position prediction accuracy along the protein sequence for three representative proteins. Each dot corresponds to one residue position; dot size is proportional to variant count at that position; color indicates local functional rate (blue = functional, red = damaging). The black curve shows a sliding window average (window = $L/20$ residues). Overall per-protein accuracy is reported in the subtitle of each panel.

Protein Fitness Prediction

Protein fitness prediction estimates how sequence changes affect a protein’s function, stability, or performance in a biological setting.

See all 20 papers

Sep 30, 2026

StabilityArc: Decoding Protein Sequence Embeddings into Generalizable Stability Landscapes

Every protein has a unique stability landscape, but the physical consequences of mutation are governed by recurring biochemical constraints. We test whether a shared decoder, trained on measurements from diverse proteins, can interpret these constraints in an unseen target, enabling cross-protein transfer for initial experimental round prescreening. We present StabilityArc , which maps frozen ESMC-600M residue representations through a shared RoPE transformer to an Lx20 matrix of substitution effects; a symmetric, contact-aware residual aids in predicting epistasis in simultaneous substitutions. In 66 strict leave-one-protein-out evaluations covering 134,794 ProteinGym variants, StabilityArc achieves 0.7134 Spearman correlation, exceeding the strongest zero-shot baseline, ProSST-2048 (0.6526), by 0.0608. We further explore the utility of this method by providing the score as a prior for Kermut, achieving Spearman correlation of 0.8280 across three supervised split schemes, improving on Kermut's reported 0.8167.

Aaron L. Feller, Andrew D. Ellington, Claus O. Wilke

The University of Texas at Austin, Interdisciplinary Life Sciences · The University of Texas at Austin, Integrative Biology

Transfer Learning Molecular Property Prediction Protein Representation Learning

Ask PDF

Sep 28, 2026

A General Harness for Protein Foundation Model Fitness Prediction

Accurate fitness prediction is central to protein engineering and understanding sequence-function relationships. With advances in deep learning, protein foundation models (PFMs) have become widely used for this task. Recent analyses, however, show that these models share preferences reflecting their training corpora, while unreliable inputs can further distort fitness predictions. Family-specific evolutionary evidence and structural context can help address these limitations by providing complementary constraints on model scores, motivating VenusREM-Harness (VRH), a general, model-agnostic, training-free Retrieval-Enhanced Mutation harness. It fuses frozen model scores with multiple sequence alignment (MSA) evidence according to model uncertainty, then applies gated background correction and score shrinkage based on structural confidence and solvent exposure. Across 1,211 assays and 3.1 million measured variants from ProteinGym, VenusMutHub, and the newly curated viral benchmark VenusViroHub, all 71 configurations improve Spearman correlation on all 3 benchmarks by 0.073 on average, with broad gains across 5 metrics. Extended analyses relate retrieval gains to model-MSA preference differences, assess domain-level gains and immune-escape cases, and quantify computational speedups. Built with VRH, VenusREM2 is the first to rank highest in all function, taxon, MSA-depth, and mutation-depth categories, with a ProteinGym Average Spearman of 0.556, 0.038 above the prior best.

Yang Tan, Qijia Tian, Gangyu Sun +5

Shanghai Jiao Tong University · Shanghai Innovation Institute · Shanghai Artificial Intelligence Laboratory

Protein Fitness Prediction Protein Language Models Molecular Property Prediction

Ask PDF

Oct 22, 2026

Trion-1: Unifying Language, Protein Sequence, and Structure with Diffusion Language Models

We envision a general-purpose protein foundation model as a unified, interactive model of sequence, structure, and function, capable of understanding, generation, and cross-modal translation while retaining the instruction-following and reasoning capabilities of modern LLMs. Toward this vision, we introduce Trion-1, the first native, unified, multimodal diffusion language model (dLLM) for proteins. Trion-1 is built on two key design choices: (1) a unified discrete token space that jointly represents sequence, structure, and natural language; (2) a dLLM backbone that replaces strict left-to-right factorization with bidirectional or semi-bidirectional token interactions, enabling global residue communication and long-range contact modeling that are naturally suited to structure modeling. This formulation unifies six core capabilities: folding, inverse folding, sequence understanding, structure understanding, sequence design, and structure design, while retaining the native language capabilities of general-purpose LLMs. On established benchmarks, Trion-1 achieves broadly competitive performance. In folding, Trion-1 achieves a TM-score above 0.8 with a single sample under strict homology control, matching or exceeding the existing tri-modal model using 200 samples. For sequence understanding, Trion-1 performs on par with top-tier LLMs such as Claude Opus 4.7 (62.37%). These results support well-aligned sequence-structure-language representations and strong cross-modal generative capabilities.

Tianyu Lu, Kaicheng Yang, Jiangtao Feng +7

Protein Language Models Protein Structure Generation Diffusion Language Models Protein Representation Learning Multimodal Diffusion Models Protein Structure Prediction+1

Ask PDF

Abstract

Protein Fitness Prediction

Explore similar work

StabilityArc: Decoding Protein Sequence Embeddings into Generalizable Stability Landscapes

A General Harness for Protein Foundation Model Fitness Prediction

Trion-1: Unifying Language, Protein Sequence, and Structure with Diffusion Language Models