Machine Learning in Pharmaceuticals, and Genomics Research

Introduction

Artificial Intelligence and Machine Learning have become common mechanisms used in Pharmaceuticals, Healthcare and Genomics research to analyze large stores of data (Big Data). As intelligent algorithms parse these Big Data stores using Natural Language Processing (NLP), Sentiment and Cognitive API’s making use of AI and ML, scientists in the Pharmaceutical, Healthcare and Genomics industry hope to identify and improve the care administered to patients with any number of clinical or behavioral health conditions.

According to an article published by Tech Emergence, they have classified 7-Applications of AI/ML as follows:

  1. Disease Identification/Diagnosis
  2. Personalized Treatment/Behavioral Modification
  3. Drug Discovery/Manufacturing
  4. Clinical Trial Research
  5. Radiology and Radiotherapy
  6. Smart Electronic Health Records
  7. Epidemic Outbreak Prediction

(TechEmergence, 2017)

While there are several potential uses of AI/ML used in Research and Development, I hope to focus on Cancer Research and Drug Discovery and Manufacturing.

Disease Identification/Diagnosis in Cancer Research

IBM Watson Health and Quest Diagnostics announced the launch of IBM Watson Genomics from Quest Diagnostics, a new service that helps advance precision medicine by combining cognitive computing with genomic tumor sequencing. In addition, Memorial Sloan Kettering Cancer Center announced that they would supplement IBM’s Genomic Research with access to their Oncology Database, OncoKB. OncoKB is a precision oncology knowledge base containing detailed information about the effects, treatment and implications of specific alterations to numerous cancer genes. IBM Watson Genomics extracts information from four primary databases (Variant DB, Statistical Reoccurrence, Treatment Guidelines, Scientific Literature) to calculate the type of actual and hypothetical treatments. (Quest Diagnostics, 2016)

Figure 1 – OncoKB

Information in OncoKB is organized hierarchically by gene, alteration, indication, and level of evidence. Implicit in the designation of a level of evidence for each branch is whether the biomarker is an FDA recognized standard care or considered to be an investigational method of treatment. E.g. predictive response to a drug that is FDA approved or currently being tested in clinical trials.

Figure 2 – OncoKB Hierarchy

(Chakravarty et al., 2017)

For each altered gene, Memorial Sloan Kettering has curated the biological effect, prevalence and prognostic information, as well as treatment implications. OncoKB currently contains treatment information for Level 1, Level 2, and Level 3 variants utilizing what they call their Levels of Evidence. (Memorial Sloan Kettering Cancer Center)

Figure 3 – Levels of Evidence

 

Results

To date, more than 3,000 unique mutations, fusions, and alterations in 418 cancer-associated genes have been evaluated and documented by researchers. Thus far, OncoKB has evaluated genomics events in over 5000 tumor samples across 19 different cancer types. Of those, Forty-one percent of samples harbored at least one potentially actionable alteration, of which 7.5% provided predictive analysis and clinical benefits over that of standard treatments, as seen in Table 1 below.

Table 1 – OncoKB Results

Drug Discovery/Manufacturing using Machine Learning

Machine Learning utilizes technology expertise in AI/ML and genome biology to predict what will happen within a cell when DNA is altered by genetic variation, whether its natural or therapeutic.

Utilizing Machine learning and large neural networks to analyze genomics data, an organization can Identify one or more genes responsible for a disease and can lead researchers to develop a drug that addresses the behavior of the faulty genes. The challenge is, drug discovery is an iterative process and there are several elements of trial and error. New drug compounds are usually based on test data taken from compounds already in existence and new compounds are discovered based on patterns isolated between those compounds. Machine Learning makes use of predictive methods based on pattern matching and other characteristic methods use in Machine Learning algorithms. (Deep Neural Nets as a Method for Quantitative Structure-Activity Relationships, 2015)

A San Francisco-based startup AtomWise, uses their Machine Learning product called AtomNet to streamline the initial phase of drug discovery, which involves analyzing how different molecules interact with one another – specifically, scientists need to determine which molecules will bind together and how strongly. They use the trial and error and process of elimination methodologies to analyze tens of thousands of compounds, both natural and synthetic. AtomNet, the first structure-based, deep convolutional neural network designed to predict the bioactivity of small molecules for drug discovery applications. AtomNet demonstrates how to apply the convolutional concepts of feature locality and hierarchical composition to the modeling of bioactivity and chemical interactions. (Izhar Wallach, 2015)

Figure 4 – AtomNet using IBM Watson ‘s Machine Learning Platform

Conclusion

With the partnership between IBM, Quest Diagnostics and use of OncoKB as contributed by Memorial Sloan Kettering Cancer Center, genomics sequencing and oncology diagnostic services will extend advanced AI/ML capabilities to more than half of the nation’s hospitals and physicians. To create a new drug, researchers must test tens of thousands of compounds to determine how they interact. After a substance is effective against a disease, it must perform well in three distinct phases of clinical trials and be approved by regulatory bodies like the FDA. (Atom Wise, 2017)

While Machine Learning Methodologies and algorithms are highly complex, the purpose of this article is to convey that Disease Prevention and Drug Discovery utilizing AI/ML can do in days what currently takes several months or even years. It’s estimated that, on average, one new drug coming to market can take 1,000 people, 12-15 years, and up to $1.6 billion dollars, AI/ML should reduce the number of physical resources, time and cost to validate effective compounds. (Singularity Hub, 2017)

Bibliography

Atom Wise. (2017). Retrieved from http://www.atomwise.com/

Chakravarty et al., J. P. (2017). Retrieved from http://ascopubs.org/doi/full/10.1200/PO.17.00011

Deep Neural Nets as a Method for Quantitative Structure-Activity Relationships. (2015). In J. Ma, R. P. Sheridan, A. Liaw, & G. E. Dahl.

Izhar Wallach, M. D. (2015). AtomNet: A Deep Convolutional Neural Network for Bioactivity Prediction in Structure-based Drug Discovery. Retrieved from https://arxiv.org/abs/1510.02855

Memorial Sloan Kettering Cancer Center. (n.d.). Retrieved from https://www.mskcc.org/ibm-watson-and-quest-diagnostics-launch-genomic-sequencing-service-using-data-msk

Quest Diagnostics. (2016). Retrieved from http://newsroom.questdiagnostics.com/2016-10-18-IBM-and-Quest-Diagnostics-Launch-Watson-Powered-Genomic-Sequencing-Service-to-Help-Physicians-Bring-Precision-Cancer-Treatments-to-Patients-Nationwide

Singularity Hub. (2017). Retrieved from https://singularityhub.com/2017/05/07/drug-discovery-ai-can-do-in-a-day-what-currently-takes-months/

TechEmergence. (2017, March). Retrieved from https://www.techemergence.com/applications-machine-learning-in-pharma-medicine/

 

 

3+

Users who have LIKED this post:

  • avatar