The pharmaceutical industry faces significant challenges, from identifying the right biological targets to designing safe and effective therapies. Traditionally, pharmaceutical companies have used artificial intelligence (AI) to build on existing treatments in the creation of me-too drugs. However, the next frontier is using AI to go beyond the known, identifying previously unexplored biological pathways and therapeutic modalities. By analyzing genetic data, researchers are now harnessing AI to discover drug-like molecules with the potential to revolutionize how they approach disease treatment.
Joe Kelleher, senior vice president in biology at Empress Therapeutics, has 25 years of leadership at the intersection between biology and chemistry, leading successful drug discovery efforts across the industry. He has worked at both biomedical research institutions and biotech companies, where he helped establish and grow specialized research groups and led the development of target portfolios and therapeutic programs. Recently, Kelleher helped to develop the Chemilogics™ platform at Empress Therapeutics, which utilizes genetic data, AI/machine learning (ML), and coevolution to predictably generate better starting points for drugs. Drug Discovery News spoke with Kelleher to learn about how his team leverages AI to analyze complex biological data, uncover hidden patterns, and accelerate the discovery of innovative treatments for different diseases.

Joe Kelleher is a senior vice president in biology at Empress Therapeutics.
Credit: Empress Therapeutics
Why is understanding biological pathways and genetic data crucial for developing effective drugs?
Drug discovery is a notoriously low-yield exercise, certain to fail if we start with the wrong targets, make drug candidates that are not sufficiently safe, or pick the wrong clinical indications for candidate drugs. Understanding biological pathways and genetic data can help address all these issues.
At Empress, we have a proprietary way to mine human clinical genetic data using AI and causal machine learning (ML) to find safe, effective small molecule modulators of disease-relevant human pathways that can be rapidly optimized for drug candidates. The key is to include the 170,000,000 genes encoded in the microbes that coevolved within our bodies (our “other genome”) in the analysis. This seeks clusters of bacterial genes that produce small molecules active on human targets in ways that cause us to be healthy.
We then use synthetic biology for the initial identification and characterization of these small molecules and follow this with chemical synthesis and optimization to derive drug candidates. Our thesis is that these will be safe and effective medicines, necessarily acting on relevant targets or pathways, and directed at their optimal clinical indications.
What are the advantages of sourcing small-molecule medicines from microbial DNA within the human body?
We see three main advantages.
First, small molecule drugs based closely on compounds already present in healthy humans, are more likely to be well tolerated and safe compared to small molecules from more common library screening (including virtual) and optimization approaches.
Second, since our ChemilogicsTM discovery engine is anchored in human clinical data, enriching only the molecules that are likely to be driving health, we believe that we are working on the most relevant and effective drug targets and pathways.
Finally, our engine builds models comparing genetic and clinical data from healthy individuals to those with diseases like inflammatory bowel disease, rheumatoid arthritis, cancer, metabolic disease, and neurological conditions, ensuring drug candidates are derived from and applied to relevant clinical indications. We see this as a drug discovery trifecta – the right molecule for the right target to help people with a specific disease.
How is your team at Empress leveraging AI to uncover previously unknown biological pathways and therapeutic modalities from genetic data?
We are not necessarily discovering unknown pathways on the human biology side, although many of the microbial synthetic pathways can be new discoveries. Perhaps more importantly, our AI/ML models lead us to human targets and pathways, allowing us to make unexpected connections to human diseases and clinical indications that were not previously appreciated. There are “aha” moments, as we find targets and pathways that we thought we knew, doing unexpected and important things in the context of a particular disease, in ways that make sense and are exciting to translate clinically.
A nice example of this is when we discovered a therapeutic mechanism where it is necessary to engage two targets with overlapping, compensatory biological effects. Like so many disease targets, industry and academic efforts previously uncovered important roles for each of the targets individually but have yet to translate these to meaningful clinical success. By following the initial, genetic-based clues from the coevolution between microbial chemists and human biological pathways, we uncovered a unique therapeutic hypothesis. I do not think we would have uncovered this through other methods. I am excited to see how this discovery translates for patients, as it may be the first example of a medicine sourced in the context of their disease.
AI is constantly evolving, bearing strengths but also bringing challenges. Can you explain these in terms of applying AI to uncover biological pathways and therapeutic modalities?
Current AI and ML methods can find patterns in data and make connections and predictions that may not be possible to find or make otherwise. However, the methods are limited by the quality and quantity of the underlying data. For drug discovery, greater amounts and quality of data, in particular, human clinical data, can help.
Our AI/ML models grow even better as we incorporate different data modalities. This includes multiomic profiling of samples from humans as hosts, paired with samples from our resident microbes, and overlay clinical features for more and more individuals, representing an increasing diversity of biology.
Are there any potential ethical or regulatory considerations arising from using AI to uncover unknown biological and therapeutic targets?
We share and take seriously our ethical obligation to find safe and effective new medicines for individuals living with diseases and view AI as a tool to help enable this endeavor.
As with so many drug discovery efforts, when patient samples are involved, protecting privacy is paramount. Fortunately, mechanisms to anonymize data are plentiful, allowing companies to correlate genomic and sample analysis data with clinical tests and health outcomes. We safeguard human clinical data used in our AI/ML models in compliance with the Health Insurance Portability and Accountability Act and other applicable regulations. We also evaluate the safety and efficacy of our drug candidates in strict compliance with FDA and other regulatory requirements.
What do you hope the future looks like for utilizing AI in the pharmaceutical industry?
I hope AI and ML methods continue to be developed and refined to help discover safe and effective therapies. Finding the right drugs for the right targets and pathways to address specific diseases remains paramount. I am hopeful that complementary methods to what we use at Empress will also work, in particular for diseases where human-resident microbes may not have a strong influence.
This interview has been condensed and edited for clarity.