A.I. and digital technologies are revolutionising genomic medicine
Medicine has always walked the fine line between reasonable prudence and a desire to innovate and adopt new technologies that have the potential to improve patient’s quality of life. Today, one of these innovations is revolutionizing the fundamentals of the medical approach: personalized medicine that aims to adapt treatments to the characteristics of each patient. This approach can be summarized as delivering; “the right treatment at the right time for each patient”. Genomics has the potential to become a key tool in the emerging discipline of personalised medicine for two reasons: its ability to reveal the factors that make each patient unique at the most fundamental level and its direct role in many rare diseases and cancers.
Potential impact of genomics on certain types of cancer
Today, the use of genomics is facilitated by technological developments known as Next Generation Sequencing (NGS) that have drastically reduced sequencing costs, from 2 billion euros 20 years ago to a few hundreds of euros today. Thanks to this evolution, doctors can access an unprecedented level of information on their patients which in many cases makes it possible to better treat or prevent a disease in a given patient.
Yet despite this huge potential, the use of genomics remains relatively limited. An obstacle to adoption that can be solved thanks to digital technologies, and in particular those of artificial intelligence (AI) and big data.
The challenge of analysing huge volumes of genomic data
Sequencing the genome of a single patient can result in more than 500 GB of information; the equivalent of a hard disk on a personal computer. Once sequenced, it is necessary to analyse this vast body of data by comparing it to a reference genome to identify and separate potentially pathogenic mutations from normal genetic differences between individuals.
Among these millions of mutations present in any individual, the challenge is then to identify the few mutations that are responsible for the disease from which the patient suffers. It is from there that one can deduce the information that may be useful to the doctor to establish his diagnosis and decide on a treatment strategy.
In addition, this whole process must be carried out in compliance with obligations of confidentiality and security of the patient’s sensitive data but in a short enough time that the patient can benefit from the results in a timely manner.
Historically, the problem of coping with these large data volumes has been addressed by excluding or filtering out all mutations that don’t have obvious links to the patient’s illness. However this approach presents a serious drawback as many diseases are influenced by several genes that can be eliminated by these filtering approaches. By removing them, it becomes impossible to see the systematic presence of a mutation considered irrelevant when it could lead to new discoveries and appropriate treatment.
SeqOne makes genomic analysis more accessible thanks to big data and AI
To address these challenges, SeqOne has developed a decision support tool, aimed at biologists and physicians, that simplifies genomic analysis while making it more accessible and faster. This solution is accessible in the cloud via a simple internet browser. The technical challenge of preserving all of a patient’s genomic information (Whole Genome Sequencing) has been made possible thanks to big data technologies (Cassandra, Spark, Elastic Search).
It is the AI that then makes it possible to analyse the volume of mutations and identify those that are relevant to a patient’s illness. For this, the solution of SeqOne considers not only the mutations detected in the patient but also contextual metadata (age, sex, symptoms, …) as well as all the information on each mutation known to the medical community. This “random forest” model of AI also learns from the user’s previous analysis, allowing for continuous improvement in performance and analysis.
Although AI is essential for the effective use of genomics in medicine, it is important to rely on “deterministic” models that provides insights into the steps that led to a given result. The alternative would be a “non-deterministic” “black box” model that provides answers without any form of justification. For ethical reasons, SeqOne remains convinced that in the field of medicine, the AI must remain firmly in the determinist camp to allow doctors to remain the final arbiter of any decisions that must be taken regarding the patient.
Towards a real precision medicine: future perspectives
The next challenge will be to analyse data from a large number of people in order to improve the learning models and ultimately improve the insights that can be provided to doctors. Thus, by analysing a growing pool of patient data augmented with information on real-world outcomes, it will be possible to precisely establish diagnoses and identify the therapy that will be best adapted to each patient.
This challenge of scaling up to increase the number of patients analysed can be solved thanks to another big data technique, integrated into the SeqOne architecture; “edge computing”. Unlike a “pure SaaS” system fully hosted in the cloud, SeqOne offers the possibility of adding a decentralized computing module that is closest to data production with a sequencer that converts biological samples into digital data. This option both automates the import of raw data, reduces data transfer times, and minimizes sensitive data that travels over the network. The result is a faster result, better network utilization optimization and increased security.
While sharing patient data has the potential to deliver better answers in treating even the most serious diseases, it also raises enormous issues regarding patient confidentiality and privacy. Indeed, the genomic data contains such a large amount of information about a patient that it would expose them to certain risks. How then can we ensure that a patient’s genomic data will not be used to discriminate on the basis of medical condition? A new approach to managing genomic data will need to be put in place. We strongly feel that governments that have already demonstrated their effectiveness in putting in place the GDPR regulatory framework for personal information are best positioned to assume this responsibility
It is by finding the right compromise and the right answers to these questions that a new era of medicine, personalized and precise, is opening up to us.
Jean-Marc HOLDER
C.O.O. SeqOne
Translated from the French. Originally published in Revue des Telecoms, issue 193, June 2019 under the title “Les technologies numériques démocratisent l’analyse génomique”