The experiment that settles it

What you get

• Experimental confirmation or refutation of the target hypothesis - both are results
• Expression confirmation in the disease-relevant tissue and matrix
• Knockdown or perturbation phenotype data
• A pre-registered analysis plan agreed before the experiment runs
• Full protocols and parameter records, written for transfer to your team

What the platform contributes

• Multi-evidence prioritisation with published weights

  Genetic association 0.30 · known-drug precedent 0.18 · tractability 0.16 · pathway 0.12 · expression 0.10 · literature 0.08 · animal model 0.06. Every score itemises its factor contributions and traces to source records, and the weights are tunable per therapeutic strategy - a rare-disease programme should not weight genetics the way an oncology programme does. You can see what we used, and you can change it.

• Formal two-sample Mendelian randomisation, before you commit bench budget

  cis-eQTL instruments as exposure against disease GWAS as outcome, estimated by Wald ratio, IVW, MR-Egger and weighted median, with Cochran’s Q for heterogeneity and the Egger intercept for directional pleiotropy. A causal question gets a causal method, not a correlation dressed up as one.

• Tractability tiering, read against known safety events

  Clinically validated (≥0.9), structurally tractable (≥0.6), druggable family (≥0.3), challenging - so validation effort goes where a drug could actually follow it.

• Knowledge-graph link prediction, honestly evaluated

  node2vec embeddings with a supervised edge classifier, and test edges held out before the embedding is trained rather than after. Every predicted association returns with the evidence paths supporting it.

• Ortholog-informed model selection

  The species you validate in is a data decision. We make it explicitly, on percent identity and homology type, rather than defaulting to mouse and discovering the problem in month five.

What you get

• A designed and empirically validated PCR or qPCR assay
• Analytical performance data: efficiency, linearity, limit of detection, specificity, reproducibility
• Optimised protocol and reagent specification
• A transfer package written so your team can run it without us
• Optional progression to kit format and OEM manufacture

What the platform contributes

• Transcript and target selection that keeps its assay and unit

  You amplify something that is actually present in the matrix you intend to sample. Expression evidence is never pooled across RNA and protein sources, because a transcript that is measured and a protein that is present are different claims.

• Variant-aware design constraints

  gnomAD allele frequencies and dbSNP identifiers flag common population variants in candidate primer and probe regions. An assay sitting on a 3%-frequency SNP fails intermittently, in the sample subset that carries it - the most expensive class of assay fault there is, because it is not reproducible on demand.

• ClinVar classifications with review-status star ratings

  A single-submitter call is never treated as equivalent to an expert-panel one.

• Ontology-resolved conditions

  MONDO and EFO, so the same indication reported three ways collapses to one node instead of splitting your evidence three ways.

• Span-cited literature evidence for the marker itself

  The quoted sentence and its offset in the source paper. A reviewer can check the claim in one click.

What you get

• Synthesised, sequence-verified constructs
• Vector design and cloning to your expression system
• Variant libraries and mutant panels where the programme needs them
• Sequence-verification data and full construct records

What the platform contributes

• Saturation scanning

  All 19 substitutions at a chosen position, ranked rather than listed.

• Stability scanning and deep mutational scanning

  Ranked candidate substitutions across the sequence, or across a specified range, so the panel you order is designed against a hypothesis instead of filling a plate.

• Binding-improvement suggestions at nominated positions

  Where the programme has a specific interface to improve rather than a whole protein to stabilise.

• ΔΔG estimation from first principles you can inspect

  Change in residue volume (packing strain), hydropathy (burial mismatch) and formal charge, plus proline and glycine backbone penalties. Where no structure is supplied, per-residue burial is estimated from windowed Kyte-Doolittle hydropathy.

• Codon optimisation routed to the people who own it

  Synthesis strategy and codon optimisation are executed by the synthesis laboratory on their established expression-system tooling. That is their domain expertise, and we route to it rather than duplicating it badly.

What you get

• Expressed and purified recombinant protein to agreed specification
• Expression system selection with the reasoning stated
• Purity, yield and activity data
• Constructs and protocols documented for repeat production
• Protein supplied as the reagent for your assay or structural work

What the platform contributes

• Developability liability scanning, with position and severity

  Asn deamidation, Asp isomerisation, acid-labile fragmentation, N-linked glycosylation sequons, free-thiol mispairing risk, Met and Trp oxidation and N-terminal pyroglutamate - each flagged at its residue, each rated, none of them buried in a paragraph.

• Aggregation-propensity profiling

  Windowed Kyte-Doolittle hydropathy across the sequence, with contiguous runs above threshold reported as aggregation-prone regions. Aggregation is the most common reason an expression campaign produces inclusion bodies instead of protein, and it is visible in the sequence before anyone orders a vector.

• Physicochemical and structural property calculation

  And, where a structure is available, secondary-structure assignment, binding-site prediction, per-residue confidence banding for predicted models, and energy minimisation.

What you get

• A named, reserved team of computational and laboratory scientists
• Continuous capacity rather than repeatedly re-scoped projects
• Direct access to the scientists doing the work, not an account manager relaying questions
• Your data, constructs, protocols and models held in your tenant
• Complete methodology records, versioned, so any past result can be reconstructed

What the platform contributes

• Why this model exists

  Discovery is cyclical. Optimisation, validation and assay work all run in loops, and a loop broken into separately scoped projects loses a fortnight to contracting at every turn. Reserved capacity removes that overhead - and it lets a team build up context that a series of discrete projects never can.

• Corpus snapshots

  Every score and every generated artefact records the immutable, named corpus state it was computed from. A conclusion reached in March can be reconstructed exactly in November.

• A point-in-time feature store

  One versioned source of features that every model reads from, with reads guaranteed not to see values postdating their label. The leakage guard is enforced by an explicit test that runs in validation, not by an intention recorded in a document.

• Append-only evidence

  Re-extraction at a new model version writes new rows and marks the old ones superseded. Nothing is destroyed, so nothing has to be taken on trust.