Axial Discovery - Machine learning and DNA-encoded libraries
Call for talented individuals and teams
Analysis sent out every Monday
Axial: https://linktr.ee/axialxyz
Axial partners with great founders and inventors. We invest in early-stage life sciences companies such as Appia Bio, Seranova Bio, Delix Therapeutics, Simcha Therapeutics, among others often when they are no more than an idea. We are fanatical about helping the rare inventor who is compelled to build their own enduring business. If you or someone you know has a great idea or company in life sciences, Axial would be excited to get to know you and possibly invest in your vision and company . We are excited to be in business with you - email us at info@axialvc.com
Machine learning and DNA-encoded libraries
DNA-encoded libraries (DEL) are a tool to screen billions of chemical compounds. DELs tag each compound with a unique DNA barcode (~20-base-pair DNA sequences) to enable templated synthesis and tracking. The premise of DELs was driven by the difference between small molecules and biological molecules: the latter are produced by a genetic code and the former are not. Biological molecules can replicate and be selected for; whereas, small molecules don’t have these intrinsic features. The idea of a DEL started with the idea that adding a piece of DNA to a small molecule can imbue a code onto them and allowing the application of selective forces to identify small molecules with certain features:
Version 1: large libraries of small molecules with little-to-no directed evolution; companies like X-Chem, HitGen, and Vipergen; bottleneck was sequencing capabilities
Version 2: larger libraries using directed evolution; companies like DiCE, Ensemble (from the Liu Lab), and Nuevolution (now part of Amgen); sequencing became cheaper and more powerful
Version 3 will have to focus on diversity, medicinal chemistry, and relieving the bottlenecks from pooling and designing new selection methods; sequencing is pretty cheap with new tools like machine learning and microfluidics coming in. Machine learning will be particularly useful to map out SARs more rapidly. An exciting development is screening libraries against whole cell environments and membranes going beyond soluble protein targets.
In particular, DELs are becoming useful to train machine learning models for virtual screening. DELs can profile ~10^9 small molecules (versus 10^6 for high-throughput screening methods) while there are around 10^60 possible small molecules. As a result, virtual screening using machine learning can more efficiently map out structure-activity relationships (SAR) across this chemical space. So DELs are a cheaper way to generate the SAR data for models to have increasingly stronger predictivity. The physical screening of a DEL provides positive and negative hits for a model. If DELs can make virtual screening useful, developing a drug for a given target might add another step: (1) traditional screen, (2) DEL, (3) virtual screen.
Some problems to solve to ensure that DELs help build accurate virtual screening models are:
Improving the diversity of DELs - scaffolds and structural diversity of the chemical matter. A problem of using DELs to train machine learning models is that a class of compounds that bind a target might not be in the library to begin with. Ensuring that molecules in a DEL are sufficiently unique from one another is incredibly important. The issue is that sometimes the chemistry required to make certain chemical matter might not be able to get tagged by DNA.
Better filters for virtual screens: structural, removing duplicates and molecules with multiple reactive groups, predicting ease of synthesis
Expand a DEL screen beyond competitive binding of a target to select for more features like potential toxicity and ADME. This will feed into more accurate models and virtual screens.
