Climbing the drug discovery ladder: aiding the biochemical pipeline with artificial intelligence approaches

The biopharmaceutical pipeline is the process from conceptualization and targeting of a desired and needed medication to actual deployment and commercialization. This process is thorough, expensive, uncertain and slow, spanning 10 to 15 years and costing as much as 2.8 billion dollars. To add more uncertainty, it has been estimated that only 12\% of the conceptualized compounds end up being approved by the FDA. With the advent of artificial intelligence (AI) tools, areas like Drug Response Analysis, Virtual Screening, and de novo Drug Design have seen high benefits from both classical Machine Learning approaches and Deep Learning architectures, achieving in some cases high performance solutions. However, the interpretability and the reliability of the resulting models is still questioned by the cheminformatics community, raising questions about the functionality of these models. In our work, we’ve aimed to produce solutions that are both interpretable and computationally efficient, providing frameworks for similarity-based drug response analysis (such as topological regression) or multitask frameworks for multi-target virtual screening, both with the aim of interpreting the chemical space and the target space. Our work shows important and novel improvements over the existing literature, achieving competing results with the state-of-the-art and providing more interpretability than competing frameworks.