How AI and human biology data led to a new addiction drug candidate

Choosing the right therapeutic target is one of the biggest challenges in early drug discovery, particularly for diseases driven by multiple biological pathways. Rather than focusing on a single protein, researchers are using AI to analyse human-derived biological data and identify mechanisms that may offer stronger therapeutic potential before compounds enter preclinical testing.

Opioid use disorder (OUD) is one such example. The condition involves widespread changes across neuronal signalling, neuroplasticity and reward pathways, making it difficult to address through modulation of a single molecular target. Researchers are now investigating whether AI can help identify combinations of targets that better reflect the biology of the disease.

Jayson Uffens, Chief Technology Officer, Co-founder and Chairman of GATC Health, believes AI should be used to support biological reasoning rather than simply identify statistical patterns within large datasets.

“My role sits at the intersection of AI systems design, biological modelling and real-world therapeutic development,” Uffens explained. “A lot of what I do is building that bridge between an AI platform, our scientists and the wet lab.”

Drug Target Review spoke with Uffens about how AI is being applied to target identification, why human-derived multiomic data matters and how GATC Health’s Operon platform was used to design a potential treatment for OUD.

When one target is not enough

Many drug discovery programmes begin by identifying a single therapeutic target before designing compounds to modulate it. Although this strategy remains effective for many diseases, it becomes more challenging when multiple biological pathways contribute to pathology.

For disorders affecting the central nervous system, including OUD, numerous signalling networks contribute to disease progression. Targeting a single receptor may therefore address only part of the underlying biology.

“Traditional discovery is largely trial-and-error and oriented around a single target,” he explained. “You pick a candidate target, screen against it and hope the downstream biology cooperates.”

GATC Health’s Operon platform analyses human multiomic datasets alongside biological pathways and disease mechanisms to generate mechanistic hypotheses that can then be tested experimentally.

“The simplest way I describe it is that this is about moving AI in drug discovery from pattern matching to biological reasoning,” Uffens explained.

Identifying patterns within large datasets does not necessarily explain why disease occurs. Mechanistic reasoning places those observations within known biological pathways, helping researchers understand why a particular target or pathway may be relevant before committing to further development.

Starting with human biology

Animal models remain widely used in drug discovery, but they do not always capture the complexity of human disease. As a result, researchers are incorporating human-derived biological data earlier in the discovery process.

“It grounds the discovery process in human disease biology from the very beginning, rather than relying solely on animal models or abstract in silico associations,” Uffens explained.

Using these datasets, Operon identified changes in serotonin signalling pathways associated with OUD. The analysis suggested that simultaneously modulating the 5-HT2A and 5-HT6 serotonin receptors could offer a more promising therapeutic strategy than targeting either receptor individually. This prediction was made before any compounds entered preclinical evaluation.

It grounds the discovery process in human disease biology from the very beginning, rather than relying solely on animal models or abstract in silico associations.

“Human-derived starting material does not eliminate translational risk – nothing does until you’re in patients – but it does strengthen the biological rationale for what you choose to test downstream,” he added.

Designing drugs for multiple targets

The resulting candidate, GATC-1021, was intentionally designed to modulate both the 5-HT2A and 5-HT6 receptors using a precision polypharmacology strategy. Rather than binding non-selectively to multiple receptors, the aim was to engage two biologically relevant targets at carefully balanced levels.

“The biology of OUD is not a one-target problem,” Uffens explained.

According to Uffens, both receptors contribute to addiction-related circuitry and neuroplasticity, making simultaneous modulation more biologically appropriate than targeting a single receptor alone.

The biology of OUD is not a one-target problem.

“Operon identified that coordinated modulation of multiple relevant nodes, at a specific activity ratio between them, was a more compelling design than pursuing a narrow ‘one lock, one key’ approach,” he explained.

The dual-target strategy was also intended to address another challenge associated with serotonergic drug discovery.

Activation of 5-HT2A receptors has historically been associated with hallucinogenic effects when targeted in isolation. By designing a compound that acts across both receptors, the programme aimed to achieve the desired biological effects while avoiding those unwanted properties.

“The fact that GATC-1021 shows a non-hallucinogenic profile despite engaging 5-HT2A is a key part of the differentiation story and it relates directly to the dual-target design,” he said.

Evidence from preclinical studies

Preclinical evaluation suggested that the compound reduced fentanyl self-administration in established in vivo models while also producing molecular changes associated with neuronal plasticity. Rather than acting directly on opioid receptors, GATC-1021 was designed to influence serotonergic pathways implicated in behavioural recovery.

“The reductions in fentanyl intake in the self-administration model suggest the candidate is affecting the reinforcing and relapse-relevant biology of opioid use, not just blunting symptoms superficially,” Uffens explained.

He also pointed to accompanying changes in biological markers associated with neuronal remodelling.

“On the molecular side, the increases in thin dendritic spines, together with changes in genes like Bdnf and Camk2a, give us cross-model confirmation of pathway engagement.”

These findings suggest the compound was affecting the intended biological pathways across multiple experimental models, providing additional evidence for the proposed mechanism.

Assessing efficacy alongside safety

Developing a compound for central nervous system disorders requires more than demonstrating efficacy. Safety considerations are equally important, particularly when targeting serotonin receptors that have historically been associated with hallucinogenic effects.

For GATC-1021, the preclinical safety package combined several complementary approaches. Researchers assessed locomotor activity, examined tissue pathology in animals from the intravenous self-administration studies and carried out in vitro receptor selectivity profiling.

“The key finding is that despite engaging serotonin receptors associated with psychedelic effects, GATC-1021 did not induce head-twitch responses,” Uffens explained.

He was equally careful to place those findings in context. Although the early data are encouraging, they do not demonstrate clinical safety.

“I want to be careful not to overclaim safety. We are pre-IND and the full IND-enabling safety pharmacology package is still in progress.”

Narrowing the search

AI is often associated with speed, but Uffens believes its value lies in improving decision-making earlier in drug discovery.

“The clearest way to see it is in the size of the search space,” Uffens explained.

He explained that Operon analysed human biomarker data, identified disease-relevant mechanisms, generated candidate scaffolds and refined them alongside medicinal chemists before compounds entered in vivo testing.

A conventional programme pursuing a similar receptor profile might evaluate around 50 compounds in vivo before identifying a lead candidate. In contrast, collaborators at the University of California, Irvine tested one comparator compound alongside a single AI-prioritised lead.

“Dr Fowler’s team at UCI ended up testing one known comparator and one AI-driven lead. That’s a meaningful gain in reductions of animal use and decision speed. This is a tangible way to describe what AI is doing in a real programme,” he explained.

Next steps

GATC-1021 is advancing through the preclinical stage, with IND-enabling toxicology, receptor selectivity profiling and broader safety pharmacology studies underway to support future clinical trials.

According to Uffens, material manufactured to Good Laboratory Practice (GLP) standards has already been synthesised to support these studies.

“The team’s intent is to move toward an IND filing and clinical trials,” he concluded.