I am a computational biologist and scientific workflow developer working at the intersection of AI for biology, single-cell genomics, computational neuroscience, and reproducible research software. I am currently a postdoctoral researcher at Hokkaido University, where I build, validate, document, and troubleshoot Python-based workflows for transcriptomics, spatial analysis, machine learning, and biological foundation models.

I am especially interested in making complex scientific analyses easier to test, reproduce, review, and extend. My work combines biological interpretation with explicit workflow interfaces, validation checkpoints, portable environments, contributor-facing documentation, and responsible AI-assisted development.

• Biological foundation models: Applying Geneformer to donor-aware single-cell classification, held-out evaluation, in silico perturbation, and biological interpretation.
• Scientific workflow curation: Organizing modular analyses with explicit inputs and outputs, configuration, validation checkpoints, limitations, and handoff documentation.
• AI-assisted software development: Using Codex and related tools for drafting, debugging, refactoring, and documentation with human review and scientific verification.
• Computational neuroscience: Contributing to multivariate neuroimaging research on language impairment after acute stroke using quantitative CT, voxel-based analysis, lesion mapping, and principal component analysis.
• Genomics and multi-omics: Developing reproducible analyses for single-cell, spatial, microbial, clinical, and environmental datasets.

A curated single-cell foundation-model workflow spanning donor-disjoint data preparation, Geneformer tokenization and fine-tuning, held-out evaluation, in silico perturbation, and spatial validation.

The repository documents workflow contracts, validation evidence, limitations, portable environments, migration procedures, troubleshooting, contributor guidance, and responsible AI-assisted development.

A reproducible multi-stage workflow using RepeatModeler2/RepeatMasker, STAR, TEcount, and DESeq2. The repository demonstrates explicit setup and input contracts, long-running process monitoring, recovery procedures, workflow consolidation, and separation of historical from canonical analyses.

1. Define inputs, outputs, assumptions, and success criteria.
2. Separate workflows into inspectable, restartable stages.
3. Add schema, row-count, leakage, biological-control, and runtime checks where appropriate.
4. Lock dependencies and record source revisions for reproducible execution.
5. Document setup, limitations, troubleshooting, recovery, and contributor expectations.
6. Use AI tools to accelerate development while retaining human review and verification.

Python · R · Bash · Linux · Jupyter · Git/GitHub · uv · Scanpy · pandas · NumPy · scikit-learn · Geneformer · transformer models · single-cell RNA-seq · spatial transcriptomics · computational neuroimaging · workflow validation · technical documentation · AI-assisted coding

Open to collaborating on:

1. AI and foundation models for biological and neuroscience data.
2. Reproducible scientific workflows and modular research software.
3. Single-cell and spatial transcriptomics.
4. Workflow validation, interoperability, documentation, and AI-agent instructions.

• Email: k.dauyey.bio.nu [at] gmail [dot] com