Drug Discovery with Deep Learning Implementation

Drug Discovery with Deep Learning Implementation

By Pharma Tech Outlook | Thursday, April 11, 2019

Pharmaceutical firms’ role in the health ecosystem is to ensure safe and efficient treatments to have an effective impact on the quality of life of patients. These treatments are discovered and developed for many times over 10-15 years and at a highly low success rate. They are complex and time-intensive and expensive. Due to the high cost of these capital markets, most pharmaceutical companies have proved very helpful in funding their financing. Such capital is, unfortunately not long-term. This has made the role pharmaceutical companies even complicated as they need to also meet the shareholder requirements for capital appreciation, security, and rapid returns. Despite significant advances in R&D sciences and corresponding technology, and management improvements, the number of new drugs approved for every billion dollars spent has decreased by about half every nine years since 1950.

Pharmaceutical efforts continue to be highly risky in an incomplete knowledge situation. A common risk management strategy involves a portfolio approach in which more experimental approaches lie alongside a basket of known quantities. Due to the lengthy time frames involved in the drug discovery process, it is crucial that the objective is soundly evaluated; the more information about the target and its viability, the more 'safe' one could assume. These known quantities will be subject to intense competition because the market is structured accordingly, so it is crucial for the portfolio to differentiate.

Crowdsourcing is a business model that uses the creative capabilities of external agents, and various applications have been implemented across the research, development, and clinical value chains. A few recent crowdsourcing applications in the literature include the development of predictive cytotoxicity models, algorithmic improvements to a popular GWAS approach, and the implementation of deep learning in the selection of compounds.

Faulty outcomes are susceptible to targeted and malicious work. Since crowdsourcing is paid per task, financial encouragement often leads employees to complete tasks quickly rather than well. It takes time to verify the responses, so applicants frequently rely on multiple workers completing the same work to correct errors. However, the numerous completion of each task increases time and monetary costs.

New Editions