Machine Learning for Science
For biotech founders, academic groups, independent researchers, and clinicians who need answers grounded in real computation, not vibes.
Valued by partners across biotech and clinical care
You bring the target. We design, train, and validate the model. Graph neural networks, transformer-based property prediction, multi-task ADMET architectures.
Send us your compounds. Useful for early triage, dose-setting questions, and building a defensible quantitative case before the wet-lab spend.
Architecture, methods, and validation review for groups building their own computational stack. We tell you what to keep, what to throw out, and where the silent failure modes are.
Independent assessment of computational claims in papers, pitch decks, or grant applications. For investors, programme managers, and editors who need a second pair of eyes that knows where the bodies are buried.
60 minutes, £100, paid upfront. We talk through the problem, what you have, what you need, and what's actually feasible. You leave the call with a clear picture of the work, even if you don't engage further.
Within an hour of the call ending, you receive a fixed quote with scope, deliverables, and timeline. No retainers, no surprise hourly bills.
We do the work. You get progress updates at agreed checkpoints and final deliverables in a form you can actually use. Standard projects close in two to six weeks.
Nabil Yasini-Ardekani
Yasini-Ardekani operates at the intersection between Materials Science and Machine Learning.
Pursuing a PhD in Computational Chemistry on the frontier of Machine Learning for Science, focused on accelerating materials discovery and cutting compute costs by up to 95%.
The same techniques transfer directly to drug discovery, where Dis-Solved operates today. Clients work with a real ML scientist actively contributing to chemical research, with eight years of teaching and research at UCL behind the ability to explain complex science plainly.
The literature is flooded with models, workflows, and pipelines. Most don't transfer to your problem. Yasini-Ardekani's job is figuring out which do, when to combine them, and when to build from scratch.
0.96
GAT / GCN / GraphSAGE architecture, ~7,800 curated compounds, stereochemistry preserved through the featurisation pipeline.
JPKPD
Journal of Pharmacokinetics and Pharmacodynamics. Mechanistic prodrug PK model for lisdexamfetamine with first-principles nutrition-aware absorption.
3
Multi-task CYP450 prediction, hERG cardiotoxicity classifier, MAT-based transporter model. Production-ready, validated on external sets.
RK4
Mechanistic PK-PD simulator built on RK4 integration of physiologically grounded ODEs, validated against published lisdexamfetamine kinetics.
Initial consultation. £100, 60 minutes, paid before the call. You leave with a clear picture of the work, even if you don't engage further.
Project work. Fixed quote, delivered within an hour of the call. Scope, deliverables, and timeline agreed upfront. No retainers. No hourly billing. No surprises.
Machine Learning for Science
Dis-Solved operates across two domains. Pick yours so the site shows the right thing.
Materials Science
The Materials practice is currently in research mode while the PhD work matures into a service offering. We'll let you know when bookings open.