Five domains — from quick answer to working capability.

Most biotech development needs some form of protein engineering — whether that's generating mutants with altered solubility or glycosylation, designing fusion constructs, choosing optimal peptide fragments for epitope mapping, or, increasingly, designing entirely new binders from scratch.

I work across both ends: classical structure-guided engineering, and the modern AI stack (AlphaFold for structure, RFdiffusion and related diffusion models for de-novo backbones, ProteinMPNN for sequence design, ESM-family models for scoring and likelihood). The interesting work is usually in combining them honestly — knowing where each method is reliable and where it isn't.

• Rational mutant & variant design
• Fusion & multi-domain constructs
• De novo binder design
• Epitope-focused fragment selection
• Stability & solubility engineering
• Glycoengineering & PTM analysis
• Developability & liability scans
• Freedom-to-operate aware design

Computational vaccinology has been my core practice for over a decade — from a published end-to-end workflow (Söllner et al., Immunome Research, 2010) to ongoing client projects across infectious disease and oncology. I help teams move from pathogen sequence to a defensible antigen short-list: reverse vaccinology, T-cell and B-cell epitope prediction, HLA and population coverage, strain consensus design.

For mRNA platforms I add tissue-specific bicodon optimization and UTR selection, and can help convert existing protein-format designs into properly engineered mRNA constructs — or build out the design layer of an entire proprietary platform.

• Reverse vaccinology pipelines
• T-cell & B-cell epitope prediction
• HLA & strain coverage analysis
• Population centroid & consensus antigens
• Bicodon & codon optimization
• UTR selection & engineering
• Protein → mRNA conversion
• mRNA platform design layer

Off-the-shelf models are extraordinary, but they aren't always the right answer for a specific question. When a project needs a custom predictor, a calibrated scoring function, a fine-tuned protein language model, or an evaluation harness that's actually trustworthy on your data — that's what this service is for.

This includes building agentic / LLM-based R&D workflows: tools that triage literature, draft hypothesis lists, summarise experimental results, or sit between scientists and a complex analysis pipeline. Currently a growing share of new engagements.

• Custom predictors & classifiers
• Protein language model fine-tuning
• Calibrated scoring & ranking
• Benchmarking & validation
• LLM & agentic workflows
• RAG over scientific corpora
• Active learning loops
• Model-result interpretation

A recurring theme across nearly all projects: the data you need is real, but it lives in fifteen places with five different schemas. I've repeatedly integrated sources such as PubMed abstracts, MeSH controlled vocabularies, drug and small-molecule target databases, the US patent full-text corpus and bespoke client data — either by pre-existing identifiers or by automated full-text mining — into queryable structures and knowledge graphs that downstream analysis can actually use.

For compute-heavy work (especially AI inference and training), I can also set up reproducible AWS environments — from a single GPU instance for an experiment to a properly versioned, containerised pipeline you and your team can run independently afterwards.

• Bio-data warehousing & ETL
• Knowledge graphs (genes / proteins / drugs / disease)
• PubMed & literature mining
• Patent full-text mining
• Reproducible pipeline engineering
• AWS GPU & batch compute setup
• Containerised, versioned workflows
• Handover & team training

Sometimes the most useful deliverable isn't code or a model — it's a clear, written assessment of what the literature actually says, what the IP landscape looks like, or whether a particular computational result should change your decision. I offer this both as standalone engagements and embedded within larger projects.

This also covers drug repositioning analysis using integrated protein/drug/disease networks — identifying compounds with mechanistic rationale for new indications — which has been a productive line of work across several past projects.

• Custom biomedical literature reviews
• IP background & freedom-to-operate searches
• Bioinformatic result interpretation
• Method scope & utility advice
• Drug repositioning analysis
• Network-based hypothesis generation
• Pre-clinical R&D decision support
• Innovation & grant-narrative support

A simple test for what to bring in-house: build the things you'll (1) do repeatedly, (2) that sit close to your USP, and (3) that you'd hate to depend on an outside party for. Everything else — one-offs, commodity steps, things that will never recur — is usually cheaper to outsource and move on.

And it depends on where you're going: a team optimizing for a fast acquisition may rationally buy more and build less; a team building a durable platform should own its core. The right answer follows your vision and what's efficient for it — not a rule of thumb. Helping you make that call honestly is part of the work, even when the answer is "don't build this, just buy it."