notesum.ai

Published at October 31

SFM-Protein: Integrative Co-evolutionary Pre-training for Advanced Protein Sequence Representation

q-bio.QM

cs.AI

cs.CL

cs.LG

Released Date: October 31, 2024

Authors: Liang He¹, Peiran Jin¹, Yaosen Min¹, Shufang Xie¹, Lijun Wu¹, Tao Qin¹, Xiaozhuan Liang², Kaiyuan Gao³, Yuliang Jiang⁴, Tie-Yan Liu¹

Aff.: ¹Microsoft Research AI for Science; ²Zhejiang University; ³Huazhong University of Science and Technology; ⁴Tsinghua University

Arxiv: http://arxiv.org/abs/2410.24022v1

Refer to caption

Task	GO-MF	GO-BP	GO-CC	EC
GearNet^∗	0.503 / 0.490	0.356 / 0.211	0.414 / 0.276	0.730 / 0.751
LM-GVP^∗	0.545 / 0.580	0.417 / 0.302	0.527 / 0.423	0.664 / 0.710
Multiview Contrast^∗	0.654 / 0.596	0.490 / 0.292	0.488 / 0.336	0.874 / 0.892
ESM2 (650M)	0.615 / 0.616	0.422 / 0.290	0.488 / 0.366	0.799 / 0.805
ESM2 (3B)	0.659 / 0.647	0.476 / 0.341	0.497 / 0.417	0.863 / 0.876
SFM-Protein (650M)	0.649 / 0.645	0.471 / 0.330	0.516 / 0.395	0.855 / 0.884
SFM-Protein (3B)	0.673 / 0.657	0.495 / 0.361	0.510 / 0.416	0.869 / 0.893