Target Identification

Identifying potential and druggable targets for developing new drugs is the first major step for curing a disease, which includes the understanding of the cellular, molecular, biochemical complexities associated with a specific disease. The classical hypothesis of ‘one gene, one drug, one disease’ in the drug discovery paradigm may have contributed to the low success rate in drug development. Using AI and related technologies in the target identification process allows scientists and pharmaceutical companies to really explore all the available evidence to better understand a disease and its underlying biology. Our team has focused on target identification, with growing amounts of data supporting early decision making.

Regular methods are based on high throughput omics-datasets, such as genome-wide association scans, gene expression analysis from tissue to a single cell, miRNA analysis, proteomics, metabolomics, post-translational modifications, cellular imaging, etc. Target-based and phenotypic screening are most commonly used approaches for finding and validating drug targets. Computational approaches offer novel testable hypotheses for systematic, unbiased identification of molecular targets of known drugs. With the rapid development of AI and machine learning technologies, unstructured information from omics, text and image analytics, public databases, systems biology can be integrated. Our expert team can help clients extracting knowledge from a multitude of resources and thus enabling science-based decisions.

Network-based methodologies for drug–target interactions (DTI) prediction in silico:

• Homogeneous Networks
  Bipartite local model (NetLapRLS)
  Kernelized Bayesian matrix factorization (KBMF2K)
• Heterogeneous Networks
  Inductive matrix completion (DTINet)
  Neural network-based approach (NeoDTI)

Compared to traditional 'black box' machine-learning methods, systems-based network analysis of large-scale biological networks will be more interpretable, visualizing prediction of molecular targets for known drugs.

Workflow

• Embed multi-dimensional networks.
• Generate biologically and pharmacologically relevant features.
• Apply a deep neural network algorithm for graph representations (DNGR) to learn features.
• Apply Positive-Unlabeled (PU)-matrix completion to find the best projection from the drug space onto target (protein) space
• Infer new targets for a drug ranked by geometric proximity.

Applications

• Target identification
• Drug repurposing

Our team is dedicated to integrating large-scale chemical, genomic, and phenotypic profiles with publicly available systems biology data efficiently to accelerate target identification and drug development both in the academic and industrial communities.

References

• Xiangxiang Zeng, et al. Target identification among known drugs by deep learning from heterogeneous networks. Chem. Sci., 2020,11, 1775-1797.