Bridging <em>in-vitro</em>, <em>in-silico</em> and corporate realms for pharmaceutical drug discovery

Abstract

The onset of the COVID-19 pandemic in 2020 highlighted the urgent need for efficient navigation of big data. In response, numerous workflows and algorithms for data processing, aggregation, and analysis were developed and widely shared within the scientific community to address these emerging health challenges. However, these workflows have limitations, which can compromise their effectiveness and reliability. First, many workflows suffer from insufficient documentation and poor version control, making them difficult to reproduce, validate, and adapt for broader use. Second, they are often tailored to specific communities or domains, limiting their robustness and generalizability across diverse applications. Third, these workflows prioritize decisions on scientific data while overlooking critical aspects such as market applicability. Fourth, integration challenges stem from an overreliance on single-modality data analysis, neglecting the incorporation of heterogeneous data types. This limitation undermines the ability to make comprehensive go/no-go decisions based on a more holistic understanding of the problem. Finally, most existing workflows remain predominantly in silico, with minimal or no in vitro validation. This lack of experimental translation raises concerns about their applicability in real-world scenarios. In our work, we developed reproducible and well-documented pipelines to integrate and consolidate knowledge across various domains to enhance pandemic preparedness. These pipelines contextualize knowledge through graphs constructed from data extracted via manual curation of scientific publications and experimental results. Designed with flexibility in mind, the pipelines are agnostic, allowing their application across multiple domains, including diverse therapeutic indication areas. Next, to expand the scope of the underlying data beyond research, we mined patent literature, a valuable resource capturing the marketing and commercial landscape of drug discovery. Using a tool we developed called PEMT, we identified patterns in compound and target-agnostic strategies employed in the commercial sector. Moreover, to address integration challenges between knowledge graphs and omics-based technologies, we merged knowledge graphs with transcriptomics data. This enabled us to decipher the mechanisms of action of key drug and drug-like candidates. Furthermore, we demonstrated the successful translation of in silico work into biological experiments. Specifically, we built a machine learning model to predict the antibacterial activity of compounds and validated it by testing an external library for antibacterial activity in both in silico and in vitro. Our work highlights the importance of combining diverse data modalities with biological networks to gain profound insights into the mechanisms driving drug discovery. These workflows and approaches lay a strong foundation for identifying and prioritizing optimal drug candidates, facilitating their transition from preclinical studies to clinical trials and, ultimately, to the market.