Technology

Multi-omics is the Key

Complex clinical variables such as drug response depend on multiple molecular factors, each of which can be measured via a different assay. Therefore, it is critical to have a multi-faceted view of disease entities. In addition, each high-throughput assay measures a biological modality overlapping but also distinct from other assays. For example, transcriptomics measures the gene expression patterns that may or may not be clearly associated with copy number patterns of the samples. However, due to this complementarity between assays, multi-omics studies provide a more complete picture of disease biology. We and others have shown multi-omics have more predictive power when it comes to modeling disease-related variables. The multi-omics data can be obtained using various sequencing techniques on biopsies, disease models or liquid biopsies. Our technology can produce and ingest multi-omics data such as mutations, copy numbers, gene expression, proteomics and DNA methylation, and effectively integrate multiple such datasets.

Data integration and correction using deep learning

The data from disease models and primary tumors often do not have one-to-one correspondence. In addition, other batch effects such as sequencing techniques, can affect the comparability of disease models to primary tumors. This creates insight transferability issues. The discordant datasets must be rectified to represent the same information during learning. Arcas technology enables simultaneous data integration and rectification, where the algorithm focuses on biological similarities between datasets and is not biased by technical differences. This way the data from disease models and primary tumors can be integrated thereby generating insights simultaneously from multiple data sets. This technology allows us to:

1) search and match disease models to primary tumors and vice versa,

2) augment data for small sample sizes so that we can still learn useful insights and simulate large scale cohorts from small clinical trials or pre-clinical PDX trials which are usually size n=15 or 20.

Scheme of data integration from multi-omics of disease models and primary tumors.

Improvement of survival prediction accuracy on held-out set

Accurate models for clinical variables

Arcas technology learns composite molecular patterns, termed “multi-omic fingerprints”, from multi-omics datasets. The fingerprints extracted from disease samples are predictive of clinical properties, such as subtype and survival. These fingerprints are high-order molecular features that represent pathways or gene sets that are differentially activated across samples. Using these multi-omic fingerprints, we are able to accurately model survival and refine molecular subtypes if necessary. Arcas technology works on any cancer type and improves modeling of clinical variables compared to using a single type of omics set or just using clinical variables such as age, tumor stage, smoking status and gender.

Biomarker identification and mechanism of action insights via explainable AI

Arcas technology relies on input from omics assays. The combination of different assays creates even more measurements and makes it even more challenging to find relevant biomarkers for the problem of interest. On top of these, relevant clinical information must be taken into account when learning from multi-omics datasets. All of these make finding relevant biomarkers more difficult than finding a needle in a haystack. Arcas Platform’s explainable AI capability makes biomarker discovery from multi-layered complex datasets straightforward and effortless. In case of drug response markers, these insights can be used to check mechanism of action assumptions and potential off-target effects. In addition, these biomarkers would also be useful for finding the right patient population in the next steps of the clinical trials.