Drug target assessment in days, every grade one click from its evidence
A graded, cited target assessment in about five days, weak evidence challenged first.
See one case, screen by screen ↓Software that answers the questions a discovery team must settle before starting a programme — human genetics, expression, pathway biology, tractability, safety and competition — and grades each with its evidence. Drug Target Assessment uses nine agents to research, challenge and assemble a graded, cited target dossier for the therapeutic-area lead to review and sign.
How to choose AI solutions for life sciences →
Updated
Before a programme starts, someone has to answer six questions about the target: does human genetics link it to the disease, and in which direction; is it in the cells that drive the disease; is there functional proof; can the planned modality drug it; what happens when it is taken away; and who else is already on it. Today a target discovery scientist answers them by hand — literature, public genetics portals, internal omics and screens, the competitor database, the patent landscape — in about four weeks of reading per target.
The result is a static document. Grades differ between scientists and therapeutic areas, claims are not tied to the passage that supports them, and the portfolio committee re-litigates the evidence. Genetics, safety and competitor evidence change every week, so a target put on hold for weak genetics stays on hold until someone happens to read the next genome-wide study.
Estimated split for one target, by hand and with the solution. The agents' research takes minutes; the days that remain are people reviewing, answering the critic and signing.
Nine agents plan, research six questions in parallel, challenge the evidence and re-grade when new studies land; the therapeutic-area lead signs.
Northwind Bio runs 12 targets across Immunology, Oncology and Neuroscience ahead of the Oct 22 discovery portfolio committee. Dr. Maya Chen, Head of Immunology target discovery, has RKN2 in ulcerative colitis to sign. Here is her week, screen by screen, in the working solution.
Twelve targets down the side, grouped by therapeutic area; six questions across — genetics, expression, biology, tractability, safety, competition — each graded A to D, with the overall grade and the stage at the right: 53 grades, 851 cited claims behind them. The evidence watch above the matrix already shows two changes: a third oral GPR214 antagonist entered Phase 1, and neutropenia was reported with a kinesin inhibitor related to KIF29.
“Six questions every target must answer, graded A to D. Click a grade to see its evidence.”
Clicking the RKN2 safety cell opens what is behind the grade. 212 heterozygous loss-of-function carriers show no immune deficiency or serious infection across 1,400 phenotypes, but two siblings with homozygous partial loss of function had chronic mucocutaneous candidiasis from early childhood, and knock-in mice clear oral Candida more slowly. Each claim carries its number, its evidence type and its source.
“Infection risk needs a monitoring plan.”
RKN2 — oral small molecule, built Oct 5 in 4.6 days from 61 sources and 142 claims. Section 5 of 6 asks what happens when the target is taken away, in people, in animals, in tissues. The safety liability agent wrote it with 81% confidence from four sources, and the source panel highlights the exact passage behind each claim, with a link to open it there.
The safety liability agent never lowers a liability without a cited source, and any human loss-of-function phenotype is flagged.
The critic reads each section as a sceptical committee member. On safety it argues that candidiasis is a known, monitorable effect of IL-17 pathway medicines and that the 212 carriers are healthy, so the risk looks dose-dependent — and proposes raising safety from C to B with a fungal-infection monitoring condition. Maya can accept, keep the grade as is, or ask the agent to look again. A second challenge — RKN1, the closest family member, is expressed in heart, so is there an off-target risk? — is resolved: 40-fold selectivity cited, and a cardiac selectivity check added as an open question for Priya Raman.
The critic proposes only — it never changes a grade itself. A person decides.
LRRK3 in ALS was signed on Sep 21 as Hold for more evidence: its genetics rested on one small association study, p = 4 × 10⁻⁵, not replicated. “Check for new evidence” re-runs the genetics, safety and competitor searches and reads 214 new papers and preprints. A genome-wide meta-analysis of 29,612 ALS cases and 122,656 controls replicates the locus at p = 2.8 × 10⁻⁹, and the risk allele raises LRRK3 in motor neurons. The cell flips from C to B with a “was C” tag, the citation slides in, and the overall grade moves from B− to B.
It stays a B, not an A, until fine-mapping is published — and Dr. Aiko Tanaka confirms before the committee pack uses it.
Sam Patel requested PTPN31 in atopic dermatitis, small molecule or degrader, for the Nov 19 committee. The assessment planner splits the template into 6 sections and 38 research questions; six specialist agents then research in parallel, each lane showing its sources and claims, and the scorecard fills as each grade lands.
The assessment planner sets the evidence bar for every grade and never grades a section itself.
Expression and competition come back A, tractability C — the phosphatase site is shallow and the allosteric pocket has no ligand yet. The critic raises 6 challenges, resolves 5 with the agents and leaves 1 for the owner. The assembler writes 128 cited claims from 54 sources into the assessment, overall B+, and routes it to Dr. Maya Chen.
Only cited claims appear in the assessment.
For the Oct 22 committee she compares RKN2, CDH31 and KCNT4 on one radar and a criterion-by-criterion table. RKN2 is strongest on genetics and pathway biology, watch safety; CDH31, an antibody–drug conjugate in gastric cancer, is strongest on expression and tractability; KCNT4 in focal epilepsy is strongest on genetics. One click adds them to the committee pack.
Maya chooses Advance with conditions over Advance to validation, Hold for more evidence or Stop. The open questions become the conditions — cellular selectivity against RKN1 in human cardiomyocytes, a freedom-to-operate opinion on the allosteric pocket, serum IL-22 as a pharmacodynamic biomarker, the QLS-208 Phase 1 read-out. The two open critic challenges are recorded with her signature, and because safety is graded C, Lena Ortiz co-signs after her.
“Signing as Dr. Maya Chen · Meaning: scientific approval of this assessment.”
The same portfolio, seen by the five people who carry it to the committee — what their month looked like, and what it looks like now.
Nine agents plan, research six questions in parallel, challenge the evidence and re-grade when new studies land; the therapeutic-area lead signs.
Splits a target request into the template sections and 35–45 research questions, assigns each to a specialist agent and sets the evidence that would earn an A, B, C or D.
Finds genome-wide and rare-variant evidence, fine-mapping and colocalisation, reports effect size, p-value and cohort size, and states whether loss or gain of function protects. Re-grades when a new study lands.
Places the target in tissues and cell types — disease against healthy, RNA against protein — from internal single-cell and bulk omics and the public tissue atlas.
Gathers mechanism, knockdown and knockout screens and animal-model evidence, with the model, effect size and number of donors, lines or animals, and places the target in its pathway.
Judges whether the planned modality can work — pockets, structures, ligands, screening hits, surface location and internalisation, knockdown and delivery, selectivity against the closest family members.
Asks what happens when the target is taken away — loss-of-function carriers and constraint, knockout phenotypes, essential tissues and class effects of medicines on the same pathway.
Maps competitor programmes by target and modality with phase, checks trial registries and congress abstracts, and summarises the patent landscape with chemical similarity to your series.
Challenges every section — single-source claims, animal-only evidence, RNA without protein, stale data, wrong direction of effect — and proposes grade changes and open questions.
Writes the templated assessment, scorecard, open-question list and references, and routes it to the right reviewer.
Accepts challenges and signs. The agents propose; a named person decides.
Ask about any target, grade or source — or say what to change. Every answer points to the passage it came from.
Why is RKN2 graded C on safety?
Because of infection risk. Two siblings with a partial loss-of-function variant had chronic mucocutaneous candidiasis from early childhood, and knock-in mice clear oral Candida more slowly. Against that, 212 heterozygous carriers show no immune deficiency across 1,400 phenotypes. The critic suggests B with a monitoring condition. That call is yours, and Lena Ortiz co-signs any C.
What changed on LRRK3 today?
LRRK3 genetics moved C → B this morning. A genome-wide meta-analysis of 29,612 ALS cases replicates the locus at p = 2.8 × 10⁻⁹, and the risk allele raises LRRK3 in motor neurons, so inhibition is the right direction. The overall grade rises from B− to B. It stays a B, not an A, until fine-mapping is published. Dr. Aiko Tanaka confirms before the committee pack uses it.
Compare RKN2, CDH31 and KCNT4 for the Oct 22 committee
RKN2 (B+) is strongest on genetics and biology: a protective variant with odds ratio 0.71; watch safety (C). CDH31 (B) is strongest on expression and tractability for an ADC: 74% of tumours stain; two competitors are ahead. KCNT4 (B) has genetics A from four families; safety C for cardiac conduction.
Add a biomarkers section to the assessment template
Done. The assessment template now has a seventh section, Biomarkers & translation — patient selection, pharmacodynamic markers and translational models. The pathway biology agent writes it and the critic checks it like any other section. New assessments get it from today; signed assessments are unchanged.
13 screens from the working solution, on its sample data. Pick one to see it large.
The template, how a grade is earned, which sources the agents may read and who signs are settings — every change is recorded on the audit trail.
| Setting | Default | Choose from |
|---|---|---|
| Assessment template | Six required sections | Genetics · Expression · Pathway biology · Tractability · Safety · Competition & IP, plus optional Biomarkers & translation and Clinical precedent |
| Grade A needs | Two independent human sources | Two independent human sources · One human source and one functional study · Any three sources |
| Critic strictness | Standard | Standard · Strict (also challenges every single-source claim) |
| Resolve every critic challenge before sign-off | Off — open challenges are recorded as conditions | On · Off |
| Safety C or D needs a co-signature | On — Lena Ortiz, safety scientist | Anyone on the team · On or Off |
| Re-check signed assessments for new evidence | Weekly | Weekly · Every two weeks · Monthly |
| When new evidence moves a grade | Show the move, owner confirms | Show the move, owner confirms · Hold the grade until the owner confirms |
| Preprint servers | Off | On (not peer reviewed — labelled as such) · Off |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample portfolio · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (a published case study of target prioritisation, 4 weeks by hand to 5 days; Minikel et al., Nature 2024 — 2.6× approval odds with genetic support, about 10% of clinical programmes approved; GOT-IT target-assessment recommendations, Nat Rev Drug Discov 2021). People, companies, targets, compounds and studies named on this page are fictional — characters and sample data in the working solution.
Software that answers the questions a discovery team must settle before starting a programme — human genetics, expression, pathway biology, tractability, safety and competition — and grades each with its evidence. Drug Target Assessment uses nine agents to research, challenge and assemble a graded, cited target dossier for the therapeutic-area lead to review and sign.
Each of the six sections is graded A (strong), B (good), C (weak) or D (gap or red flag) against a bar you set — by default, an A needs two independent human sources. Every grade opens the claims and source passages behind it, and the critic challenges any grade the evidence does not support.
Signed assessments are re-checked weekly for new genetics, safety and competitor evidence. When a new study moves a grade, the cell shows the move and the new citation, the owner is asked to confirm, and the grade counts for committee packs only once confirmed.
Yes. Alongside licensed literature, public genetics portals, the competitor database and the patent landscape, the agents read your internal omics datasets and screening and chemistry reports — only the sources you switch on in Settings, with licences respected.
No. Agents propose grades with their evidence, the critic only proposes changes, and a person decides. When new evidence moves a grade, the owner confirms before a committee pack uses it, and a grade changed by a person needs a written reason. The competition and IP agent never gives a legal opinion — freedom to operate goes to counsel.
Yes. The therapeutic-area lead decides each critic challenge and signs the assessment with its recommendation and conditions; the translational safety scientist co-signs any safety grade of C or D. Every action by people and agents is on the review record.
The design goal is a median of about five days from request to a signed assessment, against about four weeks by hand. In the working solution, the sample RKN2 assessment — 142 claims from 61 sources — was built in 4.6 days, with 38 minutes of agent time across the nine agents, and then went to scientific review.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your assessment template, grading rubric, sources and a few past target reviews — and it goes live once every quality gate has passed. We will walk you through it on your own targets first.
Sourced answers from papers, patents and reports in seconds, our own disagreements first.
1-min demoCompetitive Pipeline TrackingEvery rival trial, readout and approval in a sourced brief each Monday
1-min demoPreclinical Study Report AutomationA draft GLP report in hours, every number traced to raw data.
1-min demoWe'll run Drug Target Assessment on a few of your own targets and the evidence your team already holds.