DNA encoded library (DEL) is a unique affinity-based screening technology that have been showing increasing successes in the discovery of covalent inhibitors, allosteric inhibitors and PROTAC molecules. In the meanwhile, DEL is advantageous in selection throughput and short screening turn-around time, and can easily generate tens of thousands to millions of data points for multiple experimental conditions. Its big data approach has recently brought tremendous attentions among the scientists in the AI drug discovery field, where data-driven AI methods are challenged by the sparsity and costs of experimental data in drug discovery.
Last year, WuXi AppTec HitS teamed up with Schrödinger, a leading provider of virtual screening solutions, to develop methods to identify only the most promising drug-like compounds by complementing DEL screens with machine learning and computational workflows.
Using a series of advanced machine learning methods to train against the enormous amount of data produced from the DEL screen, the team built a model of compound potency to amplify the signal of active hits in the initial experimental results. Additionally, the team performed a workflow to filter based on drug-like properties using physics-based methods (Figure 1)*.
Figure 1. The workflow of DEL+AI screens
The results highlight the complementarity of high throughput screening and machine learning approaches to maximize the discovery of novel hits with the most potential combination of properties (Figure 2)*. Now, this machine learning-enhanced DEL screen is available via #DELight, a self-service kit created by WuXi AppTec HitS to unleash the potential of DEL technology in drug discovery.
Figure 2: Results showed improved combination of chemical properties from the top scoring machine learning-enhanced DEL ligand compared to original top scoring molecule from DEL screen.
As an emerging technology, the combination of DEL and AI has attracted great attention, which shed light on the hope for finally resolving the problems of data deficiency and data inconsistency in AI-based drug discovery.
Generally speaking, nearly billions of experimental data will be produced after a standard DEL screen. Therefore, the topic about how to deeply mine DEL data value and improve the generalization of the model is our main focus now. In addition to the affinity prediction and the optimization of physical and chemical properties for hit compounds, we are trying to build models with precise selectivity prediction power towards similar targets by making use of well-designed DEL screen logic and advanced machine learning models.
The predictive power of the machine learning model is expected to be better when interpolating – i.e. evaluating on molecules within the DEL chemical space. However, there is considerable potential to couple these models to physics-based virtual screening tools to uncover novel hits outside of the DEL library.
How to balance the relationship between prediction accuracy and structural novelty? How to expand the application scope of machine learning model under DEL scenario? There is no easy way to solve these problems, but constructive, consistently effort will make things different.
DELight provides a self-service platform for clients to explore the pool of DEL molecules without sharing target information.
WuXi AppTec HitS provides a DELight kit containing 50+ libraries and 15+billion compounds, buffers and a step-by-step online experimental manual. Drug developers perform affinity selection according to the manual and return the sample to WuXi AppTec HitS upon completion. WuXi AppTec HitS is responsible for analysis such as PCR amplification, qPCR determination, purification and NGS (Next Generation Sequencing).
Clients have the option to purchase the entire package of affinity data if promising hits are discovered. We also offer machine learning and AI enhanced hit finding options to explore more chemical space for DELight users through our partnership with Schrödinger. However, if no features are identified, the project can be terminated for cost and risk control.
Schrödinger is transforming the way therapeutics and materials are discovered. Schrödinger has pioneered a physics-based software platform that enables discovery of high-quality, novel molecules for drug development and materials applications more rapidly and at lower cost compared to traditional methods. The software platform is used by biopharmaceutical and industrial companies, academic institutions, and government laboratories around the world. Schrödinger’s multidisciplinary drug discovery team also leverages the software platform to advance collaborative programs and its own pipeline of novel therapeutics to address unmet medical needs.
Founded in 1990, Schrödinger has over 500 employees and is engaged with customers and collaborators in more than 70 countries. To learn more, visit www.schrodinger.com and follow us on LinkedIn and Twitter.