AI's application is a widely discussed topic and the use for it is heavily debated. One of the potential uses of AI is in the health and medicine field; a topic that our speakers: Dr. Yu Chuan (Jack) Li of Taipei Medical University (TMU) and Artur Kadurin of Insilico Taiwan are greatly familiar in.
Dr. Li: AI will Revolutionize Medicine Soon
Dr. Li stated that AI is not just machine learning, but also includes rule based inference, probabilistic inference, statistical modeling, and ad hoc scores. Machine learning itself consists of neural networks, support-vector machines, and more. Currently, much of medical data is image based which a specialist will analyze. However, this approach might no longer be necessary in the future. If diagnosis can rely more on AI, other forms of data can be used.
Startups or new entrepreneurs might be able to solve to the issues surrounding data storage, analysis, and evaluation. Dr. Li suggested startups interested in starting businesses with AI in healthcare to focus on 4 topics which are necessary to solve: medical errors, poor/ inconsistent quality, 1 size fits all approach, and prediction & prevention.
Medical errors are presumed to be caused by the higher complexity of medicine. Poor medical service means the patients do not receive the recommended care and the quality itself is lacking. The 1 size fits all approach will result in imprecision in medicine as everyone is treated with the same diagnosis disregarding their conditions, biodata, and more. Prevention is difficult because it is repetitive and slow; with no visible targets or effects, it is difficult to see the results and necessities of further actions. People generally do not have a proper understanding of probability and prevention measures have low market value because they are not covered by insurance. Coupled with prediction, prevention can be made more effective.
The possible AI solutions for healthcare problems include: background lifetime data surveillance with monitoring & actionable alerts; smart summarization & visualization for doctors; just-in-time diagnostics & treatment advice to prevent diseases worsening; and disease or adverse event prediction, prevention or early interventions.
Dr. Li closed his keynote speech by giving the conclusion that AI will revolutionize medicine soon and will profoundly change the doctors' work. AI implementation will inevitably affect the human workforce, including medical specialists; but when it will happen is still currently unclear.
Insilico: AI can Reduce Drug Development Price and Uncertainty
Mr. Kadurin stated that traditional drug development is costly and inefficient. Even with so much money being invested in it, very few new drugs are developed. The reason is because drug development will need to start from the very beginning, which is the disease itself. After finding what treatments need to be provided, there are a series of tests from preclinical validation tests to several phases of clinical trials that need to be passed before the drug can be made public. Out of the thousands of prototype drugs, probably only 1 will 3 phases of clinical trials.
AI can be applied for clinical trials to see which drug is more likely to pass through the trials. This will reduce the spending in research and development significantly. AI needs to be applied more for the drug discovery pipeline which consists of 3 steps: target identification, generation of novel small molecule leads, and predictors of clinical trial outcome.
The target identification process requires a predictive approach. In most cases, during this process the developers try to understand what happens when the treatment or molecule is administered. The accuracy can be improved with technology, such as predicting properties and outcome of using specific molecules. If the reaction to specific molecules can be known, AI can predict results for new molecules using AI or neural networks. The process can use information from available databases to compare modern molecules to find natural compounds which can exhibit similar effects. This way, clinical trial might not be necessary as the natural alternative can be immediately used to produce and market products.
In contrast, generation of novel small molecule leads requires a generative approach. The purpose of this step is to not only predict properties of the molecule, but also generate new molecules with desired properties. A lot of work in this step can be done by computers, which means manpower can be reallocated for other functions.
Predicting the clinical trial outcome happens after drug candidates are available and they are to be put to the clinical test. The purpose is to reduce sunk costs associated with failed clinical trials. If it is possible to predict whether a drug candidate will fail in 1 of the test phases, this can prevent further tests and thereby reduce costs.
InnoVEX Saloon – October 31, 2018
The InnoVEX team will host an event discussing AI and Blockchain on October 31, 2018. We will invite speakers to hold a panel discussion on its various applications. More information will be announced when available.