ConforFlux: Particle-Guided Trunk Repulsion for Diverse Protein Conformations
S. Suzuki, T. Amagasa
bioRxiv
Ph.D. Student
Graduate School of Systems and Information Engineering
University of Tsukuba
I am a Ph.D. student in Computer Science at the University of Tsukuba, supervised by Toshiyuki Amagasa at the Knowledge and Data Engineering Laboratory, where I work on conformational sampling in deep learning models of protein structure. As an undergraduate I studied under Tohru Ariizumi in the Ariizumi Laboratory, developing non-destructive, high-throughput metabolite measurement for Micro-Tom tomato mutant populations, and under Naoya Fukuda for my master's, working on genomic language models for plant genomes. I continue that work on genomic language models as a research assistant at the ROIS AI-Empowered Life Science Initiative (ALIS).
My work sits on the internal representations of biological foundation models: what they encode, and how they can be steered.
Deep generative models of protein structure and sequence, and the problem of controlling what they produce: recovering the alternative conformations a confident single prediction hides, then carrying that control into design.
The same models read from the inside: which features their representations encode, and how far those features correspond to biology we can already name. Reading them is what makes steering them deliberate.
S. Suzuki, T. Amagasa
bioRxiv
S. Suzuki, T. Amagasa
Journal of Chemical Information and Modeling, Articles ASAP
S. Suzuki, K. Horie, T. Amagasa, N. Fukuda
Plant Molecular Biology, 115(4), 100
Yokohama, Japan
Poster · Busan, Korea
Tokyo, Japan
Poster / Workshop · Sagamihara, Japan
Hands-on / Poster · Nagano, Japan
Poster · Hawaii, USA
Poster / Forum · Yokohama, Japan
Oral · Shizuoka, Japan
The 48th Annual Meeting of the Molecular Biology Society of Japan
Top 3% (51 selected from 1,599 entries)
Adaptyv Bio
Top 10 out of 130+ participants worldwide
158th place out of 4,189 competitors
Top 3%