Machine learning today is being used in the medical field quite regularly. Having an exact pinpoint of how our genome as a whole is evolving has been a process that has alluded genetic scientists over generations. As almost every genome requires almost 3,000,000,000 sequences of nucleotides is very difficult for the average researcher to discover patterns within the signals that may be a direct indication of a new evolutionary pressure.
Geneticists today are now using machine learning and deep learning to use algorithms that can create predictive assumptions about natural selection. Through statistical models, it is possible to automate the evolutionary inferences that we can find within our genome. Is very likely that this will be the predictive method that will become standard for geneticists in the future.
Harvard and MIT created a partnership to create a program called Deep Sweep. This program has been responsible for flagging over 20,000 nucleotides for potential further genetic study. Each one of these nucleotides could form a simple mutation that could change conditions of life for humans. These are the types of nucleotides that we could influence to help us live longer and radically in a series of diseases.
Geneticists have had mathematical models in mind for sequencing the human genome and predicting evolutionary behavior since the 1970s. Natural selection in DNA has been scientifically explained since this time. The base model showcases that if the mutation renders a person better able to survive, their offspring will continue to thrive with this new gene variant which will eventually reflect in the surplus population over time.
A great example of a mutation that we have picked up in modern times is the ability to drink cows milk and produce lactase into adulthood. This was a mutation that started in Europe thousands of years ago, and it helped to produce more healthy children, eventually becoming a norm in nearly 80% of individuals of European descent.
Training these deep learning algorithms does take time, however, an extensive amount of simulated data. Deep learning algorithms working on these patterns often operated their black box so that they can continue to develop their criteria identifying patterns within the data. To test the algorithm for the first time, the creators honed in on the natural selection genome for drinking cows milk and eventually got the program to recognize this in a variety of different studies easily.
Today researchers are sifting through data from over 1000 sequenced genomes, and it’s performing this week by a region. With the 20,000 single mutations that it picked out researchers are going to be investigating on what each one of these mutations would do edited into the DNA of living cells. The results of this study could advance a broader range of human evolution and health.
Deep learning remains an indispensable tool for sequencing genomes as well as scanning through results in detecting the signs of evolution. Pinpointing variance and recognizing areas for future research can all be done as this technology improves. It’s exciting to see how the future of genetics may be changed as a result of deep learning!