A Yorkshire-based company has developed an AI platform that can predict drug toxicity across humans and animals, with the technology designed to account for differences including sex and ethnicity.
The Royal Society of Chemistry (RSC) has called the new technology "game-changing," naming it as a finalist in the association's prestigious 2026 Emerging Technologies Competition.
The AI-powered platform, DioScor, lets medical researchers predict the toxicity of new compounds in humans and animals in just 90 seconds.
The process shows how factors such as genetics, ethnicity and age can affect a drug's effectiveness.
Predicting toxicity in 90 seconds
Speaking exclusively to Manufacturing Chemist, ConsoneAI founder and CEO Mobeen Kosar explained how the technology works.
"DioScor predicts the toxicity of a compound (including analogues, metabolites and impurities) from its structure alone."
"The user enters a SMILES string and chooses a route of administration ... within about 90 seconds, [they] get a toxicity profile across seven species: human, dog, rat, mouse, rabbit, guinea pig and hamster."
Kosar added that for each species, DioScor could predict organ-level toxicity and can also provide predictions relating to tumourigenicity.
The platform works by integrating machine learning models that are trained on historical toxicity data, utilising separate models for each species and organ.
It also incorporates structure-activity analysis and mechanistic biomarkers.
Currently, the platform can support small molecules, peptides, metabolites, impurities and payloads, as well as linkers and payload-linkers for ADCs.
The platform also provides predicted Lethal Dose 50 (LD50) and No Observed Adverse Effect Level (NOAEL) values, which can help researchers establish an initial safety margin and inform dose selection.
For humans, it predicts LD50 and NOAEL, which lets teams see an estimated safety margin at their intended dose rather than just a yes/no flag.
"Every prediction comes with a confidence score, so users can see how much weight to give each result. DioScor also shows which part of the molecule is driving a toxicity prediction, indicates the likely mechanism and relevant biomarkers and provides mechanistic insights based [on] sex and ethnicity."
AI platform recognised by RSC
Speaking on the recognition from the RSC, Kosar said: “My team and I were extremely proud to be finalists in the RSC’s Emerging Technologies Competition; it was incredible to receive validation for the value of our work."
It’s taking too long for essential drugs to get to market and with our 90-second technology, our aim is to save as many lives as possible while reducing the dangerous gender and racial bias in drug research.
Ben Voysey, Entrepreneurship Manager at the Royal Society of Chemistry, added: “ConsoneAI’s game-changing AI platform is not only an impressive technological innovation, but it also strongly aligns with the RSC’s commitment to overcoming systemic barriers and inequalities in modern-day science."
We look forward to seeing this groundbreaking concept grow from strength to strength, with the potential to transform and improve lives on a global scale.
Speaking to why she and her company developed the pioneering tech, Kosar said her drive to tackle gaps in clinical research stemmed from a deeply personal place, with three generations of her own family having experienced adverse drug-related side effects first-hand.
Global drug development has traditionally relied on a “one-size-fits-all” approach, which has often excluded women and underrepresented ethnic communities from clinical trials.
These research gaps have left the pharmaceutical industry without a clear understanding of how treatments affect different groups; thus, the platform's capacity to predict how new drugs will work across such a wide range of variables could help ensure new medicines are designed to be safe for everyone from the outset.
Targeting smarter, not just fewer, animal studies
ConsoneAI also sees the technology reducing and refining the industry’s use of animal testing.
Earlier this year, the MHRA published guidance to phase out the use of animal testing, backing non-animal methods and offering early data reviews.
If DioScor predicts a compound is likely toxic, developers can drop it before it reaches animal testing or use cell models or organ-on-chip systems first.
Kosar explains that the tech also helps determine which animal species actually need to be studied. "Predictions across seven species help teams justify which rodent and non-rodent species are relevant to humans, instead of choosing by convention."
A liver finding in rats can be read alongside the predicted dog and human results before a dog study is commissioned. Avoiding one unnecessary non-rodent study makes a real difference to animal numbers.
The potential benefit is not limited to reducing animal numbers. Kosar also argues that DioScor could help refine studies by identifying potential toxicity earlier and allowing researchers to focus subsequent animal testing on the most relevant endpoints.
"We also see a large benefit in refinement," Kosar says. In the company's ADC case study, an early prediction of liver toxicity could potentially have allowed researchers to conduct a more focused liver study in primates instead of a broader repeat-dose toxicology programme.
The total number of animals would have been similar, but the data from each animal would have been far more useful.
The technology isn't being positioned as a complete replacement for animal testing, though. Regulatory toxicology packages still require various animal studies, although regulators are increasingly incorporating New Approach Methodologies (NAMs).
DioScor is therefore positioned as a tool to reduce unnecessary, poorly targeted, or incorrectly designed studies, as well as to support exploratory work around studies that regulators still require.
Bringing toxicity testing upstream
DioScor can be used at three stages of drug development: hit-to-lead, lead optimisation and preclinical planning.
In hit-to-lead, researchers can screen potential compounds before synthesising them. This could identify toxicity problems in a chemical scaffold early, rather than discovering them months later after a compound has already been developed based primarily on potency.
In lead optimisation, chemists can assess toxicity alongside potency as they develop new analogues, impurities and metabolites.
Importantly, DioScor can indicate which part of the molecule is driving the predicted toxicity, giving chemists information they can use to modify the compound rather than simply flagging it as potentially toxic.
The platform's human and multi-species predictions can also help teams decide which animal species to use, what dose ranges to investigate and which studies to prioritise in preclinical planning.
The idea is to avoid doing an expensive GLP study when a cheaper, earlier assay could have identified a problem and stopped the programme.
From AI platform to pharma pilots
DioScor says the platform has already been evaluated by both pharma and biotech companies for applications ranging from early toxicity screening and species selection to investment decisions.
"One user estimated that DioScor could reduce carcinogenicity preclinical costs from £4m to £900,000 and shorten timelines from 24 months to 6, once NAMs become more accepted," said Kosar.
With its second-generation platform now live, the company said that it is pursuing paid pilots with pharma, CRO and drug-discovery organisations and exploring integration with preclinical software providers.
