There are many tools that could potentially inform patient care in a precise manner. The most important at the moment in terms of what it can contribute to the patient’s health in a precision medicine realm is arguably genome sequencing. The usage of genomic data to ascertain a patient’s genetic risk to develop a disease or their ability to react to a drug or even their genetic fingerprint to allow a tailored stratification of their treatment are great ways in which genomics today has a role in delivering precision medicine to patients.
Nevertheless, other than genome sequencing, there are up and coming technologies that also use high throughput data generation in the same way as genomics, which will be incorporated into the precision medicine clinic. These other -omics technologies include epigenomics (useful for example to predict genetic age), microbiome (to understand gut-health interactions), proteomics (the concentration of proteins within a sample) or metabolomics (the ascertainment of all metabolites in a sample). All of these -omic technologies will complement the clinician’s mechanistic view of the patient’s expression of health or disease.
The fact that -omics technologies have catalysed the implementation of precision medicine does not preclude however the the existence of other big data tools that will most likely impact precision medicine. A number of wearable sensors that track our vital constants, activities or dynamic measurements of metabolite levels will contribute to delivering precision medicine through the provision of potentially vast amounts of data in real time and dynamically. These sensors will be connected to apps in our smartphone that will constantly notify our health care provider to inform their decision.
The other crucial element that will also aid in the implementation of precision medicine is the field of image processing. Examples of this include the characterisation of tumour tissue, for instance. There have been some limited success stories where pathological tissues have been segregated from healthy ones and I recommend an article by Green at al, (Opportunities and obstacles for deep learning in biology and medicine), where they mention such successes.
As the above article says, and I agree with it, the potential for application of deep learning into the precision medicine realm still remains to be fulfilled. The complexity of the data, our ability to categorise them in meaningful ways as well as their availability, given their sensitivity and potential ethical usages, makes it challenging the development of the promise of precision medicine in full.
Where I see the opportunity in terms of new scenarios for supporting precision medicine is in the integration of electronic health records, sequencing data information, wearable device data and perhaps imaging (such as the one coming from MRI) to prevent disease. For that we will have to have our own health data cloud. Health data clouds are a concept that was originally presented by Leroy Hood and colleagues, and I think it has quite a lot of potential, assuming that we have the appropriate infrastructure to take care of patient privacy issues while allowing them to share their data in a controlled manner. If deep learning is able to cope with the possibility of such heterogeneous datasets (I would expect complex neural network representations for this) together with clearly delineated questions for which these datasets can be trained, then we have a chance to master the next way of precision medicine up to its promise.
That said, the low hanging fruit for precision medicine will come from the field of pharmacogenetics. We already are able to understand the metaboliser status of patients for certain drugs given their genetics. This is only going to get better. And if with the patient’s genetics we are able to add the profiling context to stratify him/her so that he/she can be enrolled in the most appropriate clinical trial for a particular drug being researched, this is going to accelerate our ability to put new drugs into the market faster or to repurpose existing ones for new useful applications quickly.
Both scenarios of diagnosis and treatment are going to be dramatically affected by the richness of new molecular and image data about the patient. Initially we will have separate data silos (e.g., genetics silos, imaging silos, electronic health record silos) which will be used independently to help improve clinical decision making (e.g., diagnosis of uncharacterised rare diseases, stratification of patient for a particular treatment). As the appropriate infrastructure to safely integrate those silos begin to arise, the power of deep learning will be significantly enhanced.
There is still a lot of work to do, however. First and foremost, we still have a poor understanding of most molecular processes as well as the functioning of the cell in its environment (after all, cells are the basic unit of life). We are just beginning to understand how cells interact and respond to their environment in a wholistic and systems level, and when we perform many molecular measurements we are just looking at averages. Single cell procedures for high throughput measuring of -omic data is also exploding as a field right now and will help us have a much more fine-grained resolution of such mechanistic processes.
My hope is that governments, industry and other organisations will be able to come up soon with legal, ethical and societal frameworks that boost the incentives for more innovation in this promising field. These will also have a tremendous impact in the future scenarios we are likely to see in regard to deep learning and big data supporting precision medicine.