Development of an analytical-predictive model based on the integration of artificial intelligence and transcriptomic data applied to the identification of molecular targets of (bio)pharmaceutical interest

Abstract

The global increase in the incidence of chronic non-communicable diseases has encouraged the pharmaceutical industry to invest in the research and development (R&D) of new drugs capable of preventing or treating these diseases. However, the development of a drug is a time-consuming and costly process. In addition, about 90% of investigational products in development fail during clinical trials because they have safety problems or lack information about the mechanism of action. In this context, transcriptomics has allowed the identification of molecular targets of therapeutic interest, predicting the mechanism of action, and even predicting eventual adverse effects. However, the results from transcriptomics have produced large volumes of data, which end up being underutilized or used inappropriately. This is because: (i) in general, RNA sequencing data (RNA-Seq) tend not to be correlated with clinical variables and; (ii) when correlated, by not using appropriate methods, they end up bringing spurious correlations that negatively impact drug R&D. In addition, (iii) due to the multidimensionality of these data, the use of conventional statistical techniques sometimes leads to results of little biological relevance. In this sense, the application of Data Analytics techniques, including artificial intelligence (AI), to integrate clinical results (metadata) with transcriptomic data emerges as a promising proposal to overcome the challenges brought in the era of Big Data. In this sense, pharmaceutical companies have invested in the development of smarter analytical strategies that allow extracting valuable information from large volumes of data accumulated over the course of the R&D of new drugs. For these reasons, this proposal aims to investigate the feasibility of an analytical-predictive model (here called BDAseq) of academic and industrial interest that, by integrating different Data Analytics and AI tools, will be able to identify therapeutic targets of relevance. For the development of BDAseq, this study will use public RNA-Seq data from areas of the brain (post-mortem) of patients with Huntington's disease (HD, model disease for the study of neurodegenerative disorders) and neurologically healthy individuals not carrying the HD associated mutation. The RNA-Seq data used in this study will be selected based on propensity score combination techniques, which will allow the most appropriate samples to be eliminated based on relevant variables contained in the clinical metadata. The raw reading matrix of these samples will be subjected to differential expression analysis by seven different methods. The results of these methods will be combined using ensemble techniques. The most differentially expressed genes will be correlated to clinical variables of interest using supervised machine learning techniques to identify the target genes These targets will be compared to the GDEs already identified by published studies. The results obtained will allow to test the viability of the BDAseq model, bringing perspectives to commercialize the model in the form of a service or even result in the development of a user-friendly software. (AU)