By Dr Neil Taylor
Artificial intelligence is everywhere in drug discovery right now. Generative models, large language models, and foundation models trained on billions of data points. The industry is awash in tools that promise to transform how we find and develop medicines. And many of them are genuinely impressive.
But they can hallucinate. They generate outputs that are plausible, confident, but quite significantly wrong in key places. In a domain where a single missed variable or an incorrectly modelled sequence or binding site can derail years of work and millions of dollars, that matters enormously.
Proasis doesn’t hallucinate. Here’s why.
At its core, Proasis works in ways that will feel familiar to anyone who understands modern AI. It analyses structural biology and protein structure data, identifies patterns across protein-ligand and protein-protein interactions, makes predictions, and surfaces insights that are genuinely novel. The outputs are generative in the truest sense – new ways of seeing relationships that weren’t exposed before.
But the foundation is completely different.
Proasis isn’t trained on vast, indiscriminate datasets scraped from the open web or assembled from disparate, unvalidated sources. Every insight, every pattern, every rule encoded in the platform has been built through 25 years of active collaboration with some of the world’s leading structural biologists, computational chemists, and medicinal chemists. The knowledge isn’t sourced from volume. It’s sourced from expertise.
There’s a meaningful distinction between having access to data, and understanding what it means and where it is useful. Large-scale AI models are extraordinarily good at the former. They can ingest enormous corpora and produce statistically coherent outputs at remarkable speed. What they often lack is grounded scientific judgement – the kind that comes from years of hands-on work interpreting crystallographic data, troubleshooting assay results, and understanding why a particular scaffold behaves the way it does in a given target context.
That judgement is what’s embedded in Proasis. The platform reflects not what is statistically probable across a broad dataset, but what is scientifically defensible according to the best experimental evidence available, continuously refined by the people who generate and interpret it.
A quarter-century of deep domain collaboration is unusual in software. Most platforms are built by technologists who learn the science. Proasis has been built with scientists, by scientists – iteratively, through sustained partnerships with structural biology, computational chemistry, and medicinal chemistry teams at leading pharmaceutical organisations globally.
That relationship changes what the software knows and how it knows it. The structural insights encoded in Proasis reflect not just published data, but the kind of tacit, hard-won knowledge that rarely appears in papers. The patterns that experienced practitioners recognise immediately, that take years to acquire, and that are nearly impossible to extract from raw data alone.
The broader AI moment we’re living through is one of scale: more parameters, more data, more compute. The assumption is that at sufficient scale, systems become reliably capable. For some tasks, that holds. For structure-based drug discovery, it’s a risky bet.
The structural biology underlying drug discovery is not a domain where more data automatically means better answers. The quality of the experimental data matters enormously. The curation matters. Fine-grained details matter. The interpretive framework matters, and the scientific context – which target, which chemotype, which therapeutic hypothesis – matters most of all.
Proasis is built for this reality. It is not trying to be general. It is trying to be succinct, accurate, and precise in the specific domain where being right determines whether a drug programme advances or fails.
One of the more underappreciated problems with generative AI in scientific applications is that these systems often don’t know what they don’t know. They produce answers with similar confidence regardless of whether the underlying evidence is strong or weak, well represented or sparse.
Proasis is different because it is explicit about the provenance of its insights. When the platform surfaces a pattern or makes a recommendation, that output is traceable to real experimental data and validated scientific reasoning. It is not a statistical artefact. It’s a conclusion.
In drug discovery, that distinction is everything.
As AI continues to transform the life sciences, the question researchers and teams need to ask is not just “what can this system do?” but “what does this system actually know, and how does it know it?”
For 25 years, Proasis has had a clear answer to that question.
For more insights into AI in drug discovery, computational chemistry and research-grade scientific software, follow follow Dr Neil Taylor on LinkedIn. Or, if you’d like to arrange a demonstration of DesertSci’s Proasis, please get in touch with our team.
