Artificial Intelligence (AI), an advancement of machine learning refers to the ability of computers to learn from existing big data. The pharmaceutical industry is a frontrunner beneficiary in the application of AI for drug R&D. AI can be used effectively in different stages of drug discovery, including rational drug design, chemical synthesis, optimization and safety prediction, drug screening, polypharmacology, drug repurposing, and drug development (clinical trials, personalized medicines and post-market surveillance). Exploring integrated multi-omics and linkage data for unique-pattern recognition using AI remains at the heart of a streamlined, automated approach to drug discovery.

The process of bringing an approved drug to the market takes approximately 13.5 years and costs approximately US$ 2 billion, (BIO, QLS advisors, 2021). This average cost estimate includes the cost for all of the development failures. A greater percentage of investment failures in the pharmaceutical industry is spent on repeated attrition of drug candidates during early development and preclinical testing. Major causes for attrition include poor pharmacokinetics (39%), lack of efficacy (30%), and toxicity (11%), (Nam et.al, Biotechnology and Bioprocess Engineering, 2020).  These are important considerations during the initial screening for HITs in preclinical development before entering clinical trials. AI-enabled drug design and development, drug repositioning and clinical trials offer a hope to significantly cut down on this cost and shorten timelines. Here we provide a short overview of different areas in drug development getting disrupted by AI.

1. AI-assisted de novo Drug Design

Novel drug target identification can be achieved with the use of ‘Deep Learning’ strategies of AI, like Natural Language Processing (NLP). It aids in automated data mining and curation from public data domains and repositories to establish semantic relationships. AI is required for trawling large volumes of multi-dimensional, heterogeneous data to predict innovative associations between drugs and targets. It also allows quicker scans through the existing literature to alleviate repetitive research tasks and find blank spots to help determine the direction of research.

Once probable targets are identified, AI can further assist in the virtual screening of compounds and predictive models for quicker validation. Numerous in-silico methods for virtual screening of compounds based on their physicochemical properties, structure and ligand-interactions allow efficient analysis, faster elimination of non-lead compounds with less expenditure. Drug design algorithms, such as coulomb matrices and molecular fingerprint recognition which consider the physical, chemical, and toxicological profiles of compounds are used in selecting the lead compound.

AI can even incorporate statistical models that can be trained to predict physicochemical properties and target binding characteristics of a compound even before synthesis. Alternatively convolution of phenotypic screens where the model predicts the protein that is likely to be targeted by the compound. These predictions will allow researchers to evaluate and validate the idea for a compound before deciding on its synthesis. Current AI can be used to invent novel molecules that hit redefined properties like conformation, solubility, lack of toxicity, and target binding.

2.  AI-assisted Drug Repurposing

Drug repurposing is exploring new indications for existing drugs (either marketed or discontinued) as an alternative over de novo drug development. Repurposing of existing drugs, including those discontinued during clinical trials for reasons other than safety, qualifies for phase II clinical trials thereby reducing expenditure and time taken than the development of new drugs. Read also ‘How artificial intelligence changed the drug repositioning paradigm?’ AI offers strategies for trawling large volumes of multi-dimensional, heterogeneous scientific data to forecast innovative associations between drugs and targets. The ability of AI to predict drug-target interactions is important in assisting the repurposing of existing drugs towards desired/intended polypharmacology. 

Anthelmintic, ‘Mebendazole’ which was identified to inhibit microtubule assembly was selected as a lead drug to target tumors. AI informed that it acts by increasing G2/M mitotic block leading to enhanced apoptosis. In vivo studies showed potency against tumor growth. AI has thus assisted in the successful repositioning of Mebendazole to treat a brain tumor. It has also been shown to mediate cell killing in vitro and in vivo in synergy with Docetaxel to treat invasive prostate cancer. Looking at the differences of brain and prostate cancer the conclusion must be that there is much more possible with this drug, but where to find the scientific justification to start further POC in animals with the right tumor models? AI can help!

‘Natural Language processing’ aids in establishing semantic relationships between for example, genes associated with a medical condition or compounds that have been shown to target those genes. Such established ‘Network Effect’ or ‘Fishing Net Effect’ containing interconnected relationships and biological entities enables identification of other active ingredients influencing similar and related signaling pathways, gene expression activities, drug target proteins, and biomarkers. This can be exploited by researchers for the identification of other targets, hidden patterns and prediction of DTIs for repositioning existing drugs, developing multiple-target and synergistic drug combinations for rare and complex indications like cancer as being carried out at Keystonemab.

3. In Clinical Trials

Generating actionable patterns by mining big data illuminates opportunities for more efficient clinical trial protocol design and management. It aids in accurate identification of suitable target patients and population size, depending on the regulatory environment and their timely recruitment thereby cutting down on the massive cost and time frame of recruitment. AI allows intelligence in many aspects of protocol design including a collection of patient-generated data through wearable gadgets and mobile applications with integrated machine learning analysis which are also used to improve patient participation, data quality, and operational efficiency in clinical trials. Prediction of site performance, analysis monitoring of historical data, pre-emptive monitoring of trial risks also helps in improving efficiency. ‘Centralized monitoring’ of trials for drug adherence, close monitoring of patients during the trial period, maintenance of records  and even ‘site-less’ or virtual trials are enabled with the use of AI. Keystonemab has developed an AI-platform containing information from 98% of investigational/failed/marketed drugs-biomarkers-signalling pathway relations. 

Keystonemab can help to de-risk your development process by expanding indications or identify plausible synergistic combinations for your molecule for pronounced efficacy. So contact us via our website for a completely free (no strings attached) call to evaluate your options and chances for co-operation.