The field of RNA therapeutics has been rapidly gaining momentum. In parallel, RNA-seq data is becoming more widely available but its utility is often limited for gene expression studies. Envisagenics utilizes its AI-driven discovery platform, SpliceCore®, to extract additional value from RNA-sequencing data using a combination of machine learning algorithms to accelerate the development of highly specific therapeutics against diseases caused by splicing errors. In 8 months, Envisagenics discovered a disease specific variant in triple-negative breast cancer (TNBC) and developed an RNA-oligonucleotide using the SpliceCore platform.

The focus on personalized medicine has had a proliferative effect on the utilization of new technologies and methodologies in the area of clinical trials. In particular, biosensors and wearable devices have proven the ability to provide complimentary data to the type traditionally collected in clinical trials and allows for a greater focus on the individual patient. A case study is presented examining such data and the application of AI/ML.

The complexity of cancer can only be understood by utilizing more data modalities. However, more data means more problems with data integration. Deep learning-based methods can sift through these complexities and efficiently stratify cancer patients based on their molecular profiles, resulting in predictive abilities for clinical variables such as survival and drug response.

With continuing advances in multi-omic analysis, including single cell and spatial transcriptomic approaches, comes the need to understand, integrate and interpret increasingly large and complex datasets originating from different technology platforms. I will discuss some of the recent strategies and trends in the fields of single cell and spatial analysis aimed at improving integration and utilization of these rich datasets, including those leveraged in our lab.

Current AI drug development approaches often only integrate a single data type which leads to a limited mechanistic view. Here, we present OneThree’s biology-driven AI discovery pipeline, which integrates 40+ distinct data types to build high performing, interpretable models. This approach significantly improves model performance and enables mechanistic interpretability. This approach has been validated in fields such as new target or biomarker discovery, indication selection, and adverse event prediction. 

Molecular technologies in combination with deep clinical phenotyping allow us to elucidate important factors involved in disease, leading to identifying new therapeutics. We previously developed a computational approach to predict novel therapeutics based on molecular signatures in drug-disease pairs and have applied it to discover new therapies from existing drugs for a number of conditions including cancer, IBD, dermatomyositis, preterm birth and most recently COVID-19, which I will present today.

AI-based technologies may change how we phenotype, in both research and clinical settings. Precision phenotyping will be discussed, particularly in the context of image-based phenotypes. Imaging is both information-rich and essential in medicine but complex to acquire and interpret. We present a use case for fetal heart disease, highlighting the use of ensemble models, small datasets and real-world clinical images as a foundational step in AI-assisted medical image interpretation.