Seungeon Lee's portfolio site — AI-driven medium-sized peptide drug discovery 2026-03-09T00:00:00+09:00 https://flansma.github.io/ Seungeon Lee lee.seungeon.62y@st.kyoto-u.ac.jp https://flansma.github.io/talks/2026-03-09-yokawa-chatgpt/ 2026-03-09T00:00:00+09:00 2026-03-09T00:00:00+09:00

Invited lecture at Yokawa Hospital (Hyogo) for ~60-70 staff covering practical LLM adoption for healthcare workflows, custom GPTs, and information security.

<h2 id="overview">Overview</h2> <p>Invited lecture at Yokawa Hospital (医療法人社団 敬命会 吉川病院) in Miki, Hyogo, for approximately 60–70 staff members. Systematically covered the full scope of AI adoption for healthcare organizations: from LLM fundamentals and prompt engineering to department-level workflow integration, organizational deployment via Custom GPTs, and information security policy design. Delivered in Japanese.</p> <h2 id="topics">Topics</h2> <ul> <li>LLM operating principles and output characteristics (hallucination, context window constraints)</li> <li>Prompt engineering: role assignment, output format specification, structured constraints</li> <li>Department-level workflow integration: administration (automated minutes formatting, email drafting, internal notice generation), nursing/clinical (simplifying patient IC materials, English literature summarization, conference abstract structuring), education/planning (training material creation, multilingual support)</li> <li>Live demo: multi-tone generation of meeting minutes and internal announcements</li> <li>Mapping AI application points across the full operational workflow (phased adoption approach)</li> <li>Context management and knowledge base construction via Projects</li> <li>Quality standardization across the organization through Custom GPTs design and distribution</li> <li>Information security: personal/sensitive data handling policies, data control settings, output approval workflows</li> </ul> <h2 id="qa">Q&amp;A</h2> <p>Also discussed application of other AI tools including CLOVA Note in healthcare settings.</p> https://flansma.github.io/publications/2026-01-10-cpi-federated/ 2026-01-10T00:00:00+09:00 2026-01-10T00:00:00+09:00

Empowering federated learning for robust compound–protein interaction prediction across heterogeneous cross-pharma domains — Journal of Cheminformatics

https://flansma.github.io/publications/2025-12-30-helm-bert-jcim-submitted/ 2025-12-30T00:00:00+09:00 2025-12-30T00:00:00+09:00

HELM-BERT: A Transformer for Medium-sized Peptide Property Prediction — Journal of Chemical Information and Modeling (JCIM)

https://flansma.github.io/preprints/2025-12-29-helm-bert/ 2025-12-29T00:00:00+09:00 2025-12-29T00:00:00+09:00

HELM-BERT: A Transformer for Medium-sized Peptide Property Prediction — arXiv

https://flansma.github.io/talks/2025-10-29-helm-bert-cbi/ 2025-10-29T00:00:00+09:00 2025-10-29T00:00:00+09:00

Oral presentation (Award winner) at CBI 2025 introducing HELM-BERT, a transformer for medium-sized peptide property prediction using HELM notation.

<h2 id="overview">Overview</h2> <p>Oral presentation at the CBI Society Annual Meeting 2025 on HELM-BERT, a transformer-based language model designed for medium-sized peptide property prediction using HELM notation. This presentation received the <strong>Oral Presentation Award</strong>.</p> <p><strong>Session:</strong> O03-02</p> <p><strong>Authors:</strong> Seungeon Lee, Takuto Koyama, Itsuki Maeda, Atsuyuki Matsumoto, Yasushi Okuno</p> <p><strong>Affiliation:</strong> Kyoto University Graduate School of Medicine, Department of Biomedical Data Intelligence</p> https://flansma.github.io/projects/helm-bert/ 2025-01-01T00:00:00+09:00 2025-01-01T00:00:00+09:00

Transformer-based language model for medium-sized peptide property prediction using HELM notation.

<h2 id="overview">Overview</h2> <p>HELM-BERT is a transformer-based language model designed specifically for medium-sized peptides represented in HELM (Hierarchical Editing Language for Macromolecules) notation.</p> <h2 id="key-features">Key Features</h2> <ul> <li>Pre-trained on large-scale peptide databases</li> <li>Fine-tunable for various property prediction tasks</li> <li>Supports non-natural amino acids and modifications</li> </ul> <h2 id="results">Results</h2> <p>The model achieves state-of-the-art performance on peptide property prediction benchmarks.</p>