The rise of Machine Learning in drug discovery
Are we experiencing the new trend in artificial intelligence ? Let’s find out within 7 min of reading !
Introduction
How long do you think it takes for a new drug to get approved in the US? About 12–15 years, that’s roughly 1/3rd duration of the entire professional career for any of us. Next, how much investment do you think is required to bring a new drug to market? ~$1 Billion, that’s equivalent to employing 665+ employees at the annual salary of $100k for 15 years.
Simply put, approval of a new drug is a massive undertaking and selecting the right partners, who prioritize in sync with you and focus on high quality and fast turnaround, goes a long way in avoiding missteps and moving your drug program closer to success.
Can you all still blame scientists for not inventing life saving drugs as quickly as possible ?
Let’s quicky see what phases are involved in drug discovery !
Phases
Understanding the phases involved in complete drug discovery is essential because machine learning is all about first understanding the problem statement. This article focuses more on understanding machine learning behind the drug development rather than developing the market ready drugs LOL!
- Step 1: Discovery and Development
- Step 2: Preclinical Research
- Step 3: Clinical Development
- Step 4: FDA Review
- Step 5: FDA Post-market Safety Monitoring.
Major Machine learning algorithms in Drug discovery
1. Random Forest (RF)
RF is a widely used algorithm explicitly designed for large datasets with multiple features, as its implifies by removing outliers, as well as classify and designate datasets based on relative features classified for the particular algorithm.
In drug discovery, RFs are mainly used either for feature selections, classifiers, or regression utilized RF methods to improve affinity prediction between ligand and the protein by virtual screening through selecting molecular descriptors, based on a training data set for enzymes.
2. Naive Bayesian (NB)
NB algorithms are a subset of supervised learning methods that have become an essential tool used in predictive modeling classification.
Classification of biomedical data is crucial in the drug discovery process, especially in the target discovery subset. NB algorithms have shown great promise as classification tools for biomedical data, often filled with non-related information and data, known as noise .NB techniques could also serve important roles in predicting ligand-target interactions, which could be a massive step forward in lead discovery.
3. Support Vector Machine (SVM)
SVMs are supervised machine learning algorithms used in drug discovery to separate classes of compounds based on the feature selector by deriving a hyper plane.
SVM is crucial to drug discovery because of its capability of distinguishing between active and inactive compounds, ranking compounds from each database or training regression model. SVM can be attributed in various scenarios. SVM classification has a subset binary class prediction that could differentiate between active from inactive molecules.
AI can tailor approaches for a more accurate understanding of pathological cellular and molecular mechanisms.
Sounds extravagant huh? Let’s deep dive into it !
CASE STUDY 1.
BenevolentAI : USING MACHINE LEARNING TO IMPROVE TARGET PREDICTIONS
The Company BenevolentAI, a UK company founded in 2013, creates and applies AI technologies to transform the way medicines are discovered, developed, tested and brought to market. The company has over 200 biologists, chemists, engineers, informaticians and data scientists working in cross-functional squads and is headquartered in London with a research facility in Cambridge (UK) and further offices in New York. BenevolentAI has active R&D drug programmes in disease areas such as ALS, Parkinson’s, ulcerative colitis and sarcopenia. It has established partnerships with a number of major biopharma companies.
The AI solution for drug discovery
BenevolentAI has the capability from early discovery right through to late-stage clinical development. The company has developed the Benevolent Platform — a leading computational and experimental discovery platform that allows their scientists to find new ways to treat disease and personalise medicines to patients. The Benevolent Platform® focuses on three key areas, Target Identification, Molecular Design and Precision Medicine.
Main projects and diseases areas
BenevolentAI’s platform produced a ranked list of potential ALS, treatments, together with biological evidence. The BenevolentAI team was able to rapidly triage these predictions using strategies focussed on pathways implicated in multiple ALS processes. The five most promising compounds were taken to the Sheffield Institute for Translational Neuroscience (SITraN), a world authority on ALS. An ALS lead molecule emerged from a breast cancer drug, which showed delay of symptom onset when tested in the gold standard disease model.
In April 2019, the company began a long-term collaboration with AstraZeneca, aimed at using AI and machine learning to develop new treatments for chronic kidney disease (CKD) and idiopathic pulmonary fibrosis (IPF).
In September 2019, BenevolentAI signed a Framework Collaboration Agreement with Novartis Pharma AG (“Novartis”). This initial project with Novartis in oncology will see the application of AI and ML technology to stratify patients and gain a better understanding of patient and disease heterogeneity to more precisely target medicines for patients who need them.
Achievements
The company aims to use the power of AI to put patients first, and tangibly transform their lives by creating a way to lower drug discovery and development costs, decrease failure rates and increase the speed at which medicines are delivered to patients. BenevolentAI has published several pieces of research in distinguished scientific journals and world-renowned conferences.
DL technology is not only transforming the small molecule research field, but it is also showing potential in the identification of new biologics.
Curious about how ? Let’s get over the case study right away !
CASE STUDY 2.
ATOMWISE
USES DL TO SPEED UP DISCOVERY AND FIND MOLECULES FOR THE HARDEST TARGETS
The Company Atomwise, a US company founded in 2012, uses AI technology to predict small molecule-protein binding affinities and focusses on identifying potential therapeutics for any disease target. The company has 46 employees. It has set up over 300 partnerships with major biopharma companies and academic research centres around the world. In 2018, it secured US$45 million in venture capital funds for further development of the AI technology, with a total of US$51.3 million in funding to date.
The AI solution for drug discovery
The AI platform AtomNet is a patented structure made of DL Convolutional Neural Networks for hit discovery and lead compound identification and optimisation. It learns the three-dimensional features of drug-to-target molecular binding and identifies discriminators. The platform can select hits that have key features such as the ability to cross the blood-brain barrier in a short amount of time with new lead compounds obtained in days, bypassing the need for costly and long high-throughput screening experiments.
Main Projects
They are working with partners across the globe on drug discovery projects for a variety of diseases, including Ebola, multiple sclerosis and leukaemia.
Achievements
The company used AI technology and algorithms in partnership with the University of Toronto for the rapid identification of treatments against the Ebola virus. The results of the research have been submitted to a peer-reviewed publication. Atomwise also found a new molecule targeting multiple sclerosis that inhibits a protein-protein interaction in the central nervous system and has been shown to be orally active in mouse models at very low dose. The drug has been licensed to an undisclosed bio-pharma company. More recent achievements include successes on Chagas disease, hand-foot-and-mouth disease, ischemic stroke and Parkinson’s disease.
The future of drug discovery: Delivering ‘4P’ medicine
The adoption of AI and other innovative technologies, and the use of big data from multiple sources is enabling more precise targeted treatments and shifting the health ecosystem toward a future where medicine is personalised, pre-dictive, preventative and participatory (the ‘4Ps’), leading to new, more efficient and effective models of care. Over the next decade, these shifts will have a significant impact on treatments and on patient outcomes, particularly in those areas of medicine with unmet need.
