I am not discounting the “researchers” in the study. I am only objecting to the forward-looking statements in the write-up.
I am emphasising the specific words:
Specifically, the technology has helped the laboratory scale up its Clinical Knowledgebase, or CKB too – a vast searchable database that helps oncologists and other healthcare experts make detailed interpretations of complex sequencing and maintain troves of leading-edge insights – drawn from thousands of cancer research papers each day – to help drive personalized treatments.
The machine learning technology, which is still evolving, is increasingly able to “read” complex medical and research documents – trained to highlight important and relevant information contained within them such as new insights into genetics, drugs and patient response.
That mining of disparate knowledge sources means clinicians can save hours finding and curating relevant data, targeted to specific genomic profiles.
No one is using the genomic sequencing and then applying it in the clinics regularly. It is an incredibly complex (and costly) proposition. However, the highlighted words only represent a “potential”. Not an actual treatment scenario. That would require extreme computing resources. Unless the company can directly link to the EMR’s and running the algorithms.
Most of the papers coming out in the biomedical domain are usually “proof-of-concepts”. The relevant methodology to sequence genes also requires a broad consensus and standardisation. The strength of the reagents and the algorithms to splice them are also an essential factor. It’s interpretation and looking for mutations in a specific genomic browser is also not for the faint-hearted. Personalised/Precision medicine is a chimaera; AI is the new snake oil to peddle forward-looking statements.
Know what? Most people only require a human touch, sympathetic ear and palliative care to achieve the goal of treatment. I am not discounting the technological advances, but it requires intense scrutiny. Input data to run algorithms needs to be structured before you can draw meaningful correlations. Unstructured data can probably point out the direction, but it is likely to be a falsified correlation. As such, we need to start from backwards- quantify outcomes before running AI.