RipplePLM: Structural and Property Decoupling for Protein Mutation Effect Generation
Abstract
Protein mutation effect generation asks a model to describe the functional consequence of a point mutation in natural language. Existing protein-to-text systems typically encode mutation information into undifferentiated representations, overlooking the organization of mutation-induced evidence across structural and biochemical factors. We propose RipplePLM, a mutation-aware generation framework centered on Direct-Distal Cross-Attention (DDCA). By constructing a residue-level Mutation Perturbation Field from pre-trained protein language models, DDCA leverages predicted contact maps to organize mutation representations into two pathways: the mutation site's immediate contact neighborhood and its multi-hop distal context. To complement this structural decomposition, we further introduce the Property Latent Chain (PLChain), which injects expert-guided supervision of biochemical property changes (e.g., thermostability and optimal pH) into the LLM hidden-state pathway through latent property tokens. On MutaDescribe, RipplePLM improves over mutation-specific baselines on temporal and structural splits; under a matched-backbone comparison, average structural-split ROUGE-L increases from 22.23 to 35.65. Expert evaluation further shows a higher proportion of biologically accurate or relevant descriptions than the mutation-specific baseline. Additional ablations, representation diagnostics, and low-N fitness regression experiments further support the effectiveness of the learned mutation-aware representations. Code: https://github.com/Lyu6PosHao/RipplePLM.
Related
- Mechanochemical Decoupling of ATP Hydrolysis and RNA Translocation in SARS-CoV-2 nsp13 by the L405D Mutation.
- Mutation-induced structural and dynamic perturbations at the IL-17A–IL-17RA interface: Therapeutic insights from multi-replicate molecular dynamics simulations
- Structural Basis of Second-Generation HIV-1 Maturation Inhibitor Activity by Magic Angle Spinning NMR
- CP-GFM-SGX: Topology-Aware Cyclic Peptide Generation and Design with Receptor-Conditioned Structural Evaluation and Experimental KRAS Affinity Optimization
- SF3b inhibitors suppress proliferation of nonsmall cell lung cancer cells irrespective of EGFR mutation status
BibTeX
@misc{lv2026rippleplm,
title = {{RipplePLM: Structural and Property Decoupling for Protein Mutation Effect Generation}},
author = {Liuzhenghao Lv and Yuyang Liu and Yuyang Gao and Li Yuan and Yonghong Tian},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.01891},
eprint = {2610.01891},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.01891},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - RipplePLM: Structural and Property Decoupling for Protein Mutation Effect Generation
AU - Liuzhenghao Lv
AU - Yuyang Liu
AU - Yuyang Gao
AU - Li Yuan
AU - Yonghong Tian
PY - 2026
DA - 2026/10/03
PB - arXiv
DO - 10.48550/arxiv.2610.01891
UR - https://arxiv.org/abs/2610.01891
AB - Protein mutation effect generation asks a model to describe the functional consequence of a point mutation in natural language. Existing protein-to-text systems typically encode mutation information into undifferentiated representations, overlooking the organization of mutation-induced evidence across structural and biochemical factors. We propose RipplePLM, a mutation-aware generation framework centered on Direct-Distal Cross-Attention (DDCA). By constructing a residue-level Mutation Perturbation Field from pre-trained protein language models, DDCA leverages predicted contact maps to organize mutation representations into two pathways: the mutation site's immediate contact neighborhood and its multi-hop distal context. To complement this structural decomposition, we further introduce the Property Latent Chain (PLChain), which injects expert-guided supervision of biochemical property changes (e.g., thermostability and optimal pH) into the LLM hidden-state pathway through latent property tokens. On MutaDescribe, RipplePLM improves over mutation-specific baselines on temporal and structural splits; under a matched-backbone comparison, average structural-split ROUGE-L increases from 22.23 to 35.65. Expert evaluation further shows a higher proportion of biologically accurate or relevant descriptions than the mutation-specific baseline. Additional ablations, representation diagnostics, and low-N fitness regression experiments further support the effectiveness of the learned mutation-aware representations. Code: https://github.com/Lyu6PosHao/RipplePLM.
ER -