Artificial Intelligence for Life Science Diagnostics
There are multiple forms of artificial intelligence (AI), including deep learning, neural networks, Bayesian networks, and evolutionary algorithms. Here’s how AI approaches are currently being applied across life science and in metabolic disease research.
Artificial Intelligence in Pharmaceutical Research
Pharma is using AI in three main ways for developing new therapeutics:
- New target identification
- Drug development
- Improving clinical trials
Identifying New Targets
AI can help identify new drug targets in multiple ways. RNA sequencing can be used to profile gene expression. Gene expression data is then used to measure the efficacy of a therapeutic intervention, to mine for new pathways, or to identify new biomarkers of disease.
DNA sequencing has two major applications: germline (inherited) and somatic mutation profiling.
Germline mutation data can be used to assess a patient’s genetic risk of developing a given disease (e.g. BRCA1, 2) or to diagnose inherited genetic diseases.
Somatic mutations - those that accumulate in a patient as they age - can be used in a variety of ways, including:
- Diagnosing cancer (e.g. FLT3, NPM1, P53)
- Tracking relapse/remission
- Acting as a non-lethal endpoint in clinical trials
- Indicating resistance to a given therapy
- Guiding clinicians in choosing the most efficacious therapy for a specific patient
AI has a massive role in the drug development process. Most major pharma companies already have AI projects or collaborations underway, such as:
- Pfizer, who are collaborating with IBM to use their Watson artificial intelligence system to power immuno-oncology drug searching and more quickly analyze and test hypotheses.
- Sanofi, who are hunting for new metabolic disease therapies through a strategic research collaboration with Exscientia, who have a unique AI platform for identifying and validating combinations of drug targets.
- Genentech, who are using an AI system from GNS Healthcare to power cancer drug development with causal machine learning.
Multiple AI companies currently also offer drug discovery, design, and repurposing platforms.
Clinical Trial Improvements
AI is used to improve clinical trials through targeted recruitment, helping the right patient find the right trial more quickly and easily. If AI can extract medical record data, specific patient disease can be compared with criteria for all suitable open trials.
AI could also help with medication and protocol adherence, which are big problems within clinical trials. Mobile technology and apps can remind patients when to take medication, but investment is also being made in “ingestible sensors” which could track drug intake and wireless pill bottles. This could allow drug companies to be more certain of data which would otherwise be self-reported by patients.
Al can also help in clinical trials with drug efficacy prediction algorithms.
AI in Pathology
AI is used in pathology to allow more objective interpretation and analysis of data. For example, immunofluorescent or chemically stained cells and tissues can be analyzed at a more objective level than the current, traditional subjective process. Some AI companies are focusing on augmenting the work of doctors through such image interpretation and prognosis predictions.
The diagnostic imaging market is one of the most mature functions for AI. Emerging applications are extending the advances made in diagnostic image analysis to other image analysis tasks, like identifying cancer cells in stained pathology slides.
Immunostaining encompasses a broad range of techniques used in histology, cell biology, and molecular biology, which all use antibody-based staining methods. Using AI to interpret chemical stains, such as H&E, could be used to mine historical pathology data, automate repetitive tasks, or add additional data to clinical trials.
Data Integration using AI
AI-supported data integration is an emerging best practice in the life science industry. Data integration has historically proved challenging, as data sets are large, often in incompatible formats, and are always growing.
Traditional approaches involved standardizing data formats and then manually scripting and querying data to generate useful data sets. By contrast, AI approaches use machine learning and natural language processing to integrate comprehensive data sets and mine them for valuable insights. This method enables scalable integration and analysis.
AI in Metabolic Disease
Changing the Way Diabetics Manage their Health
As well as helping in immuno-oncology drug hunting, IBM’s Watson is aiding diabetics in managing their health.
At the ADA 78th Scientific Sessions in June, IBM discussed its advances in using AI, machine learning, and analytic technologies to address the data-driven obstacles in diabetes management.
Through a collaboration with Medtronic, an app has been developed that aims to help people manage their diabetes. The Sugar.IQ™ diabetes assistant helped keep blood sugar levels in normal ranges for nine extra days over a year and is now commercially available for use.
Causal Machine Learning (CML) and NASH
Causal machine learning (CML) is a form of AI used to gain insight into disease progression and determine clinical outcomes after treatment. This is achieved through leveraging numerous types of clinical data to identify prognostic and predictive biomarkers.
CML is being used to help people with NASH, a lifestyle disease with few symptoms and no reliable diagnostic tests. GNS Healthcare and Gilead Sciences presented data at The International Liver Congress™ on building predictive survival models. These models predicted the onset of cirrhosis or clinical events, such as liver transplant or disease complication.
Through powerful AI mining of clinical data, genetic links that to help identify patients with NASH could be unveiled. This can lead to better understanding of which drug combinations are right for a specific patient and more reliable translation of clinical trial results into real-world outcomes.
Since the term “Artificial Intelligence” was coined in 1956 by John McCarthy at MIT, its application to human health care has supported significant progress. While there is still much more work to be done, this an exciting field to watch for future advances.