In a recent campaign, Tes Pharma achieved a hit rate exceeding 60% from a billion-scale virtual screen using deepmirror™. This result significantly outperforms typical hit rates of 5-10% for challenging transcription factor targets. 

In our previous post, we described how Tes Pharma screened over one billion molecules against a different transcription factor target in just hours using deepmirror. Here, we share the experimental validation of that effort. Despite testing only a small subset of predicted candidates, the results exceeded expectations, delivering an exceptionally high confirmation rate across a highly diverse chemical space. 

The challenge: targeting transcription factors 

This outcome is particularly notable given the inherent difficulty of transcription factor drug-discovery. Unlike classical targets, transcription factors typically lack well-defined binding pockets, making them challenging to modulate with small molecules. Successful discovery efforts therefore require:

• Exploration of vast chemical space 
• Identification of rare, non-obvious chemotypes 
• Reliance on high-signal predictive models rather than brute-force enumeration  

These constraints have historically limited the effectiveness of both structure-based and ligand-based screening approaches for this target class. 

The approach: billion-scale, uncertainty-aware screening 

The deepmirror platform enabled Tes Pharma to overcome these limitations. By automatically scaling computational infrastructure and delivering predictions with robust uncertainty estimates, the screening pipeline supported systematic prioritisation across ultra-large chemical libraries. 

This approach allowed Tes Pharma to identify novel chemical scaffolds that would have been extremely difficult to uncover using traditional methods. Structure-based approaches were constrained by limited protein structure or pocket information, while conventional ligand-based workflows were restricted by computational barriers to exploring chemical space efficiently at this scale. 

By combining large-scale inference with uncertainty-aware ranking, deepmirror enabled efficient navigation of an exceptionally broad and diverse molecular landscape. 

Key drivers of success 

These results reinforce three central factors behind the campaigns performance: 

1. The deepmirror model captured molecular features relevant for modulating the transcription factor target 
2. Uncertainty-aware prioritisation focused experimental resources on the most promising chemical series 
3. Even with a limited number of compounds tested, the virtual screen translated into real biological activity 

In practical terms, this demonstrates that billion-scale exploration was not merely computationally ambitious, but biologically meaningful, generating experimentally confirmed hits with high precision. 

Several of the identified scaffolds are now undergoing deeper evaluation, representing new entry points into target class long considered “undruggable”. 

Experimental validation and industry perspective 

Dr. Francesco Greco, Head of Computational Chemistry at Tes Pharma, commented: 

“Seeing such a strong hit rate from a billion-scale virtual screen, tested on only a small subset of compounds, confirms that AI-powered discovery is not just fast, but genuinely predictive.”

This follow-up campaign underscores a fundamental shift in virtual screening. Billion-scale exploration is now both practical and experimentally meaningful. 

With deepmirror, navigating vast chemical space no longer requires specialised infrastructure or months of computational effort. A few lines of code via the deepmirror API, combined with a robust internal dataset and a challenging target, were sufficient to generate hits with exceptional precision.

Dr. Ryan Greenhalgh, CTO and Co-Founder at deepmirror, added:

“Ultra-large scale AI-powered screening is only useful if we can measure impact. Experimentally confirming more than 60% of selected molecules as hits is strong evidence of the impact AI can deliver for drug discovery teams.”

Looking ahead 

Tes Pharma is now extending this approach across additional discovery programmes, integrating large-scale predictive modelling more deeply into its research workflow.

These programmes are supported by deepmirror’s large-scale predictive modelling platform, enabling the transition from data to hypotheses, and from hypotheses to experimentally confirmed hits, with increased speed and efficiency.

Together, this collaboration illustrates how uncertainty-aware, billion-scale AI screening can translate directly into validated chemical matter, accelerating progress against some of the most challenging targets in drug discovery.