Harnessing big 'omics' data and AI for drug discovery in hepatocellular carcinoma

Abstract

Hepatocellular carcinoma (HCC) is the most common form of primary adult liver cancer. After nearly a decade with sorafenib as the only approved treatment, multiple new agents have demonstrated efficacy in clinical trials, including the targeted therapies regorafenib, lenvatinib and cabozantinib, the anti-angiogenic antibody ramucirumab, and the immune checkpoint inhibitors nivolumab and pembrolizumab. Although these agents offer new promise to patients with HCC, the optimal choice and sequence of therapies remains unknown and without established biomarkers, and many patients do not respond to treatment. The advances and the decreasing costs of molecular measurement technologies enable profiling of HCC molecular features (such as genome, transcriptome, proteome and metabolome) at different levels, including bulk tissues, animal models and single cells. The release of such data sets to the public enhances the ability to search for information from these legacy studies and provides the opportunity to leverage them to understand HCC mechanisms, rationally develop new therapeutics and identify candidate biomarkers of treatment response. Here, we provide a comprehensive review of public data sets related to HCC and discuss how emerging artificial intelligence methods can be applied to identify new targets and drugs as well as to guide therapeutic choices for improved HCC treatment.

Fig. 1 |. Translational research and big data.

Translational research comprises four main components: patients, tissues, in vitro models (cell lines and organoids) and in vivo models. Each component can be characterized by different molecular modalities (such as genomics, epigenomics and functional genomics). Artificial intelligence (AI) can be used to improve the insights from big data by delineating differences and similarities and further facilitating efficient therapeutic discovery. CNV, copy number variation; miRNA, microRNA.

Fig. 2 |. Connecting genomic features and therapeutic targets in HCC.

Highly mutated, amplified or deleted genes were extracted mainly from The Cancer Genome Atlas (TCGA) analysis work published in 2017 and a review by Llovet et al.. Therapeutics in phase II and III trials for patients with hepatocellular carcinoma (HCC) and their targets were mainly selected from the review by Llovet et al.. Many new trials of HCC therapeutics are being launched; new therapeutics in the latest trials that do not pursue new targets were not included. The expression of these targets was retrieved from cBioPortal, and the percentage of patients with a target expression z score >2 was computed for each target (shown under Targets in trials). The column of targets in trials suggests that therapeutic targets in HCC are expressed in a small portion of patients with HCC, and the column of genomic features suggests that none of the genomic features are altered in half of the patients with HCC. Few connections between genomic features and therapeutic targets indicate the gap in translating genomic features into therapeutic targets. All drugs in clinical trials tested alone or in combination; all drugs reached phase II clinical trials unless otherwise stated. A3AR, adenosine receptor A3; CCR4, CC-chemokine receptor 4; FGF, fibroblast growth factor; HDAC, histone deacetylase; mTOR, mechanistic target of rapamycin; PD-1, programmed cell death 1; PDGF, platelet-derived growth factor; PD-L1, programmed cell death 1 ligand 1; SK2, sphingosine kinase 2; STAT3, signal transducer and activator of transcription 3; TGFβR1; TGFβ receptor 1; VEGF, vascular endothelial growth factor. *Phase III clinical trials.

Fig. 3 |. Translating big data to therapeutics.

A target-based approach involves the modulation of one single protein by small or large molecules; a systems-based approach involves the modulation of a list of disease-related molecular features (for example, gene expression, protein expression, metabolite abundance) mainly through small molecules. Other therapeutic strategies, such as immunotherapy, are not included. CNV, copy number variation; miRNA, microRNA.