2018 is probably the year of immunology. Last month, two immunologists – James Allison and Tasuku Honjo were awarded Nobel Prize in Physiology or Medicine, for their discoveries of checkpoint inhibitors which revolutionize immunotherapy.
In August, a group of researchers from France had successfully established for the very first time that artificial intelligence (AI) is able to process medical images to sift out clinical details. The team designed an algorithm to look at CT scan images and devised “radiomic signature”. It is a predictive score to illustrate whether the immunotherapy is effective for the patients based on the level of lymphocyte infiltration.
At the beginning of the year, John Hopkins’ scientists employed AI to create the ImmunoMap, a map which details cellular receptors on the exterior of T-cells which gives other scientists a clearer picture of an immune system’s responses to antigens. Eventually, the map will aid in the identification of receptors which show better responses to immunotherapy drugs.
AI on human immunity is precision
It’s believed that a larger presence of lymphocytes yields a greater chance for immunotherapy to be effective. However, there is no marker to accurately predict patients’ outcome or to improve odds of treatment. Such demand for predictive biomarkers to determine whether a patient will react to the immunotherapy gives AI the perfect platform to step into.
Dr. Sandip Patel, Associate Professor of Medicine, Co-Leaders of Experimental Therapeutics and Deputy Director of the San Diego Center for Precision Immunotherapy at the University of California told The ASCO Post that, often immunotherapy is advance cancer patients’ last line of defence, especially if palliative care is not their choice.
Despite so, only 15-20% of the patients will benefit from immunotherapy. There are drugs to stimulate the immune system to battle against cancer; they may trigger the body to start attacking normal tissues like the way to do for cancerous tissue.
Furthermore, there is no test whether patients will develop immune-related toxicities. This is something which cannot be determined based on a person’s demographics and conditions. AI zooms into precision and individuals, which fills in the gap provided by population statistics that only look into the mass.
Human on AI immunity is bias
While AI is rather capable in helping us with our immune system, it’s not very good with its very own immune system. As pointed out by Dr. Moriz Hardt, Assistant Professor from the department of electrical engineering and computer sciences of University of California, Berkeley, AI’s way of “approaching fairness” is biased towards human thinking.
For example, a luxury hotel which is often visited by wealthy white people but less by less affluent Black people, is more likely to know how to promote their latest offers to the former group of clients. Since algorithm is generated by people, the exact bias will be generated onto the AI itself.
That’s why AI will never be able to tell the exact differences between a good and a popular selfie, because computational researchers have not yet master the techniques of teaching AI social values, cues and human behaviours. It’s interesting to observe the sharp contrast between human and AI on the topic of immunity. Either all, there’s still a lot to work on.