AI + Genomics = Custom Made Medicine?!?!

6 min readJan 13, 2020

This year, millions of people will be getting some sort of disease, whether it is Alzheimer’s, cancer, diabetes, Parkinson’s disease, etc, and each person will be fundamentally, genetically different, which means each person will be susceptible to different disease, different side effects from drugs, and even how effective the drug is. However the treatment they will be getting won’t account for all of this, instead, it is a“one-size-fits-all” treatment.

Often times when developing treatments and medicines, they often base it off of the population averages. This traditional method often misses its target, because everyone’s fundamental, genetic makeup is different. This is where Personalized medicine can help.

Personalized Medicine has existed for quite some time. Over the years, physicians would slightly alter their treatment based on the information collected from the individual patient. However, physicians were only able to slightly alter treatment because they had SOME information about the individual patient.

But what if you get to know everything about the patient, like what type of disease they are most likely to get, what effect can this drug have, or how effective a drug is on the patient? Then doctors would be able to make a treatment that would be tailored for you. So then if this is possible, how can we get this information? It’s not like the patients know the answer to this. Instead, the answer lies in the code of life, genes.

🧬 Genome Sequencing

Before we look into how AI and Genomics are disrupting the industry, we need to learn about genome sequencing.

In the last two decades, there have been some crazy advancements in the field of genomics that have allowed us to sequence our genome and read some of the information from it. Genome sequencing is basically mapping out a person’s whole DNA in the body. But how do we actually “sequence a genome”? Watch the video below to get a better understanding:

Summary:

A genome is all the genes that make an organism. Genes are made up of DNA, and DNA is made of long paired strands of A — Adenine, C- Cytosine, G — Guanine, and T — Thymine.
The opposite letters bind together, to make up the DNA. A binds to T, and C binds to G
Your genome is the code that tells your cells how to behave
Process of Gene Sequencing
First, break the long pieces of DNA into smaller pieces
Each of these pieces is separated and sequenced individually
Since the genome pieces are so small, we want to amplify the signal of each letter. So to do that they add an enzyme that makes thousands of copies of each genome piece. Now we have 1000s of replicas of the genome with the same combination of A, C, T, and G
After amplifying the small pieces of genome, we need to read it. To do this we add special colored markers that bind to the opposite letters on the DNA strand. So now there is a colorful spot for each letter.
The scientists then take pictures of each snippet of the genome, and seeing the order of the color allows them to read the sequence
The smaller sequences are all stitched together by a computer algorithm so that we the complete sequence of the entire genome
By sequencing the genome does not mean that we can read the genome, scientists are working on how to interpret it.

🤔 Who cares?

Now that we’ve got a gist on how the genome is sequenced, we may wonder what do we actually learn from our genome?

Right now we can mainly interpret 3 things from our genome:

Single — gene disorders are diseases that are caused due to mutations in the DNA of one gene. Examples include fibrosis, sickle cell disease, Fragile X syndrome, muscular dystrophy, and Huntington disease
Multifactorial disorders are diseases associated with changes in the DNA in multiple genes. Our environment also has an impact on the disease. Examples include heart disease, diabetes, and obesity.
Pharmacogenomics Profile helps doctors understand how effective will certain drugs be on this patient. This is where personalized medicine comes in!

Genomic profiles of patients have also been increasingly used for risk prediction, disease diagnosis, and development of targeted therapies (1).

👉 Personalized Medicine

After we have the sequenced genome, we can use the information from it along with AI to give treatments specifically tailored towards each patient’s genetic, lifestyle, and environmental characteristics.

For precision medicine, we mainly need 2 things, really powerful computers (maybe quantum computers?), and big datasets of biological data. Algorithms identify patterns in large biological datasets and use the learnings to predict or optimize based on the availability of similar data on individual patients.

AI and the sequenced genome data can help us with many tasks that lead up to personalized medicine. It can help with the Variant calling and interpreting, and Relationship extraction. When talking about using AI for personalized medicine, it’s not necessarily always a classification algorithm for a genome sequence, it may include using NLP to look through research papers, and other biomedical literature.

Variant Calling and Interpretation

AI algorithms that are used to read through sequenced DNA to look for complex structural variants. ML models such as CNNs would be quite useful for this type of identification problem. Google’s DeepVariant has made the variant calling problem into an image recognition problem, by converting the genome data into images. Only knowing whether there is a variant or mutation is not enough, we also need what it does. Many research groups are training ML models on features encoding secondary structures, intrinsic disorders, DNA-binding, phosphorylation, conservation, predicted structure, and homolog counts to further improve the accuracy of variant classification, to incorporate high-dimensional data sets, and to unify the variant interpretation among laboratories (1).

Relationship Extraction

Relationship Extraction helps to come up with meaningful, evidence-based recommendations. Algorithms look for relationships between entities found in the biomedical literature to find associations of specified genetic alterations, conditions, and treatments. These can be used as evidence to prove genotype to phenotype data, such as an association of certain variants and drug sensitivity. Over recent years ML has played a huge role in relation extraction, and entity extraction. Automated text mining tools such as BeFree are able to identify gene-disease, drug-disease and drug-target associations with state-of-the-art performance.

So essentially what we want is to have the AI model find variants in the genome, figure out how it affects the patient and then look through biomedical literature to find associations/relationships with certain other things. This will allow clinicians to more carefully tailor individual interventions — whether it be disease preventive, or modifying — instead of the current process, which is often symptom classification and vague process of treatment decisions.

Why do we NEED Personalized Medicine?

Personalized medicine is going to completely revolutionize how we are taking drugs. We will so much more about our body, and how to take care of it more efficiently, and effectively. By looking into our genome we will figure out which diseases each person is susceptible to, and how we can take preventive steps before the disease even occurs. We will be able to get treatments that are much more effective, and less side effects. And most importantly, this will reduce time, costs and failure rates in pharmaceutical clinical trials. Personalized medicine is just a glimpse into the future where tech and humans co-exist.

Works Referred:

Hey, hey, hey! If you are reading this, thank you 🙏 🙏 for making it to the end!

I’d love to connect through LinkedIn, and learn about your thoughts on Personalized medicine!