Studies consistently find that 90 percent of clinical candidates never reach regulatory approval, and a significant share of those failures trace back to target selection rather than chemistry or formulation. Identifying a plausible target from genomic data grows more tractable with artificial intelligence (AI), yet the step that follows remains stubbornly manual. Scientists must still triage every candidate flagged by an identification platform before committing experimental resources, and as those platforms generate ever-larger output lists, that review bottleneck constrains discovery throughput.
OutSee, a Cambridge-based genomics and drug discovery company, has received a grant through Innovate UK's Growth Catalyst Investor Partnership programme to build a prototype agentic AI sub-system for automating the initial triage of target candidates produced by Nomaly, its core predictive genomics engine. The £50,000 grant is jointly provided by Innovate UK and existing investor Empirical Ventures, matched by private investment from Empirical Ventures under the programme's co-funding model.
Nomaly's hypothesis-free design sets it apart from conventional AI target discovery tools
Unlike most AI target discovery tools, which learn from annotated datasets of previously confirmed gene-disease relationships and are therefore constrained by what the existing literature has characterised, Nomaly takes a different approach: hypothesis-free genomic first-principles modelling, grounded in the fundamental molecular and cellular biology of the genome rather than prior associations. This lets Nomaly surface targets in underpowered datasets, in previously analysed cohorts that standard pipelines have exhausted, and in disease areas where genetic architecture is complex or poorly mapped.
OutSee applies Nomaly to an in-house discovery pipeline focused on neurology, rare, and metabolic diseases. In May 2026, the company received a £100,000 Discovery Award from the Longitude Prize on ALS, a five-year, £7.5 million challenge to accelerate drug discovery for amyotrophic lateral sclerosis (ALS), the most common form of motor neurone disease (MND), granting access to genomic data from 9,000 ALS patients for a nine-month target identification project. Nomaly is also deployed in partnerships with Genomics England and the Finnish Biobank Cooperative (FINBB), two major population genomics initiatives in Europe.
Because Nomaly can extract signals from datasets that other tools discard as too small or already-analysed, its candidate output volume runs ahead of the scientific team's capacity to review each one manually. Initial validation screening requires expert judgment on biological plausibility, druggability, and the strength of evidence linking a candidate to disease, and OutSee's scientists currently carry out every assessment by hand. Scaling that review step in step with platform output is not feasible without automation.
What will the agentic AI sub-system actually do?
The sub-system acts as a first-pass filter between Nomaly's output and the human validation team, assessing each candidate against biological evidence, druggability data, and disease-area relevance before a scientist reviews it. Agentic AI systems differ from standard machine learning (ML) models by reasoning across multiple steps, synthesising information from heterogeneous sources, and executing sequential decision logic without continuous human instruction. A recent review in Clinical Pharmacology and Therapeutics found that automation stalled at expert interpretation, leaving the review step intact despite gains in throughput; agentic architectures are the first class of tools with the structural potential to compress that step itself.
The aim is not to replace expert judgment on high-priority candidates, but to cut the volume of manual work on candidates the automated layer can confidently rank low, redirecting scientific time toward assessments that require deeper domain expertise.
The pilot study will build and test a prototype sub-system across five defined parameters:
- Prototype sub-system for initial triage of Nomaly's genomic target output
- Automated candidate assessment before human scientist review
- Architecture designed to scale validation throughput without proportional headcount growth
- Pilot scope: proof-of-concept study, with results to be disclosed on completion
- Funding: £50,000 Innovate UK grant with matched investment from Empirical Ventures
AI target identification produces more candidates than scientists can manually screen
Computational scores from platforms like Nomaly represent hypotheses about disease biology, not confirmed causal relationships. Conflating a high score with validated biology is the most common error at this stage; establishing that a target drives human disease rather than merely correlating with it requires functional confirmation, typically clustered regularly interspaced short palindromic repeats (CRISPR)-based perturbation in a disease-relevant cellular model.
Identification and validation cannot collapse into a single step, but triage automation determines which candidates reach the validation queue and how quickly. Drug mechanisms backed by human genetic evidence carry a 2.6-fold approval advantage over those without it, which is the core rationale for genomics-driven platforms like Nomaly. That same genetic orientation tends to surface large numbers of candidate signals, each requiring expert filtering before experimental resources are committed.
| Stage | Current OutSee workflow | Proposed workflow with agentic sub-system |
|---|---|---|
| Target identification | Nomaly AI engine; hypothesis-free genomic modelling | Unchanged |
| Initial triage | Manual review by scientific team | Agentic AI automated pre-screening |
| Expert validation review | Scientists assess shortlisted candidates | Scientists focus on pre-filtered shortlist |
| Experimental validation | CRISPR-based or other functional assays | Unchanged |
| Lead target progression | Partnered with o2h Discovery for lead candidate | Unchanged |
Julian Gough, CEO and founder of OutSee, framed the sub-system as an extension of Nomaly's existing capabilities. "By building agentic AI into our validation pipeline, we will maximise the impact of our exceptional target discovery scientists and accelerate discovery of high-quality novel targets," Gough said.
Dan Bolser, principal bioinformatician at OutSee, described the practical effect of removing the manual bottleneck. "With a tailored agentic AI sub-system in place, we'll be able to quickly scale up our validation pipeline, removing a key bottleneck by automatically triaging Nomaly's high-volume output," Bolser said. "This in turn will allow our scientific team to focus more attention on validating novel drug targets, where their expertise has the greatest impact."
The Innovate UK grant extends a pipeline already moving toward the clinic
OutSee closed a £2.5 million seed round in January 2026, completing an oversubscribed financing led by Ahren Innovation Capital with participation from Kadmos Capital, Panacea Ventures, Empirical Ventures, and 26 angel investors. That capital funds experimental validation of multiple AI-identified targets alongside continued Nomaly development, and a partnership with o2h Discovery, a preclinical contract research organisation, is advancing a drug discovery programme around the company's lead target candidate. The agentic AI pilot adds an automated triage layer to a platform that is already generating targets across neurology, rare, and metabolic disease programmes.
OutSee has not disclosed a timeline for integrating the sub-system into its live pipeline or expanding beyond the initial prototype. OutSee scoped the work as a proof-of-concept, and whether it demonstrates sufficient triage accuracy to displace manual review at scale will determine its path from prototype to production tool. The company's focus on neurology, rare, and metabolic diseases, indications with a history of target-based failure, makes automated triage particularly high-value in its pipeline.
This article is based on a press release issued by OutSee and was produced under Drug Discovery News' AI Editorial Guidelines.
