LineageFlow: Flow Matching for High-Fidelity Family-Aware Protein Sequence Generation

Langzhang Liang, Ming Yang, Feng Yi, Junfan Li, Shirui Pan, Xu Yinghui, Tianlei Ying, Yizhen Zheng, Zenglin Xu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:73652-73674, 2026.

Abstract

Protein sequence generation for engineering requires samples that are biophysically plausible and, when targeting a family/domain, remain recognizable members while exploring within-family diversity. Current discrete generative models typically start from uniform or masked-token noise, which discards strong position-specific constraints induced by evolution and forces the model to reconstruct conserved residues from scratch, leading to weak family control and low foldability. We propose LineageFlow, a simplex-valued flow-matching model that initializes generation from lineage priors derived from ancestral sequence reconstruction, turning generation into structured mutation from an evolved scaffold. Across diverse protein families, LineageFlow achieves family validity close to held-out natural sequences and improves predicted structural confidence over uniform-/mask-initialized baselines while maintaining substantial within-family novelty and diversity, even surpassing a large pretrained baseline trained on substantially more data. Finally, we introduce rerouting, a single intermediate-time mutate–select–amplify intervention that enables objective-guided sampling without per-step predictor guidance and yields further gains in plausibility, including a zero-shot enzyme generation case study.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-liang26aa, title = {{L}ineage{F}low: Flow Matching for High-Fidelity Family-Aware Protein Sequence Generation}, author = {Liang, Langzhang and Yang, Ming and Yi, Feng and Li, Junfan and Pan, Shirui and Yinghui, Xu and Ying, Tianlei and Zheng, Yizhen and Xu, Zenglin}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {73652--73674}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/liang26aa/liang26aa.pdf}, url = {https://proceedings.mlr.press/v306/liang26aa.html}, abstract = {Protein sequence generation for engineering requires samples that are biophysically plausible and, when targeting a family/domain, remain recognizable members while exploring within-family diversity. Current discrete generative models typically start from uniform or masked-token noise, which discards strong position-specific constraints induced by evolution and forces the model to reconstruct conserved residues from scratch, leading to weak family control and low foldability. We propose LineageFlow, a simplex-valued flow-matching model that initializes generation from lineage priors derived from ancestral sequence reconstruction, turning generation into structured mutation from an evolved scaffold. Across diverse protein families, LineageFlow achieves family validity close to held-out natural sequences and improves predicted structural confidence over uniform-/mask-initialized baselines while maintaining substantial within-family novelty and diversity, even surpassing a large pretrained baseline trained on substantially more data. Finally, we introduce rerouting, a single intermediate-time mutate–select–amplify intervention that enables objective-guided sampling without per-step predictor guidance and yields further gains in plausibility, including a zero-shot enzyme generation case study.} }
Endnote
%0 Conference Paper %T LineageFlow: Flow Matching for High-Fidelity Family-Aware Protein Sequence Generation %A Langzhang Liang %A Ming Yang %A Feng Yi %A Junfan Li %A Shirui Pan %A Xu Yinghui %A Tianlei Ying %A Yizhen Zheng %A Zenglin Xu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-liang26aa %I PMLR %P 73652--73674 %U https://proceedings.mlr.press/v306/liang26aa.html %V 306 %X Protein sequence generation for engineering requires samples that are biophysically plausible and, when targeting a family/domain, remain recognizable members while exploring within-family diversity. Current discrete generative models typically start from uniform or masked-token noise, which discards strong position-specific constraints induced by evolution and forces the model to reconstruct conserved residues from scratch, leading to weak family control and low foldability. We propose LineageFlow, a simplex-valued flow-matching model that initializes generation from lineage priors derived from ancestral sequence reconstruction, turning generation into structured mutation from an evolved scaffold. Across diverse protein families, LineageFlow achieves family validity close to held-out natural sequences and improves predicted structural confidence over uniform-/mask-initialized baselines while maintaining substantial within-family novelty and diversity, even surpassing a large pretrained baseline trained on substantially more data. Finally, we introduce rerouting, a single intermediate-time mutate–select–amplify intervention that enables objective-guided sampling without per-step predictor guidance and yields further gains in plausibility, including a zero-shot enzyme generation case study.
APA
Liang, L., Yang, M., Yi, F., Li, J., Pan, S., Yinghui, X., Ying, T., Zheng, Y. & Xu, Z.. (2026). LineageFlow: Flow Matching for High-Fidelity Family-Aware Protein Sequence Generation. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:73652-73674 Available from https://proceedings.mlr.press/v306/liang26aa.html.

Related Material