Artificial intelligence does all kinds of things….genomics comes to that as well.
Genetic engineering has always been a go-to plot twist in sci-fi movies and TV shows. The idea of genetically mutated humans with superior abilities and unique DNAs still has ripple effects on Marvel fans and box offices.
But what if we can alter genes in real life? CRISPR gene editing has been doing that since 2012 (no Wolverine or Magneto though). In 2022, this powerful genetic engineering technique is complemented with artificial intelligence.
By joining these forces, we can get the all-time best appearances of CRISPR and revolutionize the healthcare industry.
Today, I’ll go over the nuts and bolts of AI-powered gene engineering and demonstrate the difference machine intelligence makes, using one of my company’s projects as an example.
In layman’s terms, genome editing is a technology that enables scientists to modify the DNA of organisms, be it plants, animals, or humans.
It involves altering the DNA sequence of a gene to correct a genetic disorder or to make a change in the gene that confers resistance to disease. The development of CRISPR-based genome editing tools has revolutionized this field, making it easier and faster to modify genes.
Therefore, since 2012, gene engineering has been mostly associated with CRISPR-Cas 9 or just CRISPR, which is a unique “cut-and-paste” tool to modify genomes.
CRISPR-Cas9 has a massive application area. Thus, it can silence or activate genes, as well as remove or introduce new ones. So far, this technique has already been used to treat a variety of disorders, including cancer and infections. Also, CRISPR has been used to suppress mosquito populations to curb malaria and enhance crop climate resilience in engineering agriculture.
As for potential use cases, CRISPR fares well for minimizing genetic disorders, treating and preventing the spread of diseases, and improving crop performance. ExxonMobil also counts on gene modification to produce algae biodiesel by 2025.
But if it’s all rainbows and butterflies with CRISPR, why am I even writing this?
The answer is simple: **CRISPR Cas9 is not the savior we hope it is.
A growing body of evidence suggests that along with positive DNA changes, CRISPR is also guilty of occasional unintended genetic alterations. Thus, errant on-site edits genome editing outcomes are present in around 16%of the analyzed human cells.
It means that the off-target effect can change a gene's function and cause chromosomal instability, limiting the gene's potential and application in clinical procedures. Moreover, the editing performance of this technique can be impacted by a slew of factors and reduce the number of successfully edited cells.
And this is where genetic engineering and artificial intelligence collide.
Artificial intelligence and its off-shoots are usual suspects when it comes to trailblazing solutions and breakthroughs. Genomics is no exception. Thanks to the analytical and predictive capabilities, AI technologies save the day and lives by amplifying the editing power of CRISPR.
Let’s go over the main areas where AI chips in with meaningful patterns and algorithms.
As I pointed out above, CRISPR Cas9 is a traditional gene-editing tool that corrects gene defects, thus treating a disease. The success of this undertaking stems from targeting the right cells and the right tissue. However, the probability of unexpected cleavage prevents the en masse adoption of CRISPR and overshadows its potential.
Off-target mutations have long been a subject of research. As a result, scientists produced multiple scoring systems that evaluate target and off-target effects. However,the majority of existing off-target prediction methods cannot factor in the evolving CRISPR-Cas9 data for continuous self-learning.
Most importantly, traditional scoring tools do not analyze “the potential relationships between mismatched and matched sites, which may affect the off-target activity in CRISPR-Cas9 gene editing.”
And this is where artificial intelligence steps in.
It makes CRISPR Cas9 more accurate by predicting both the target and off-target effects with an unrivaled accuracy score (around 97%). From a technical standpoint, both deep neural networks and ML algorithms can be applied to produce predictions.
To train the model, engineers typically feed the algorithm with off-target datasets, such as CRISPOR one, for performance benchmarks. Additional evaluation is also much welcome and can be done by adding another off-target dataset.
A computational model by Microsoft is a prominent example of AI casually rescuing the accuracy of gene editing. The tool allows scientists to locate the best sites for gene editing, while also reducing the possibility of side effects.
Since some illnesses can be caused by one’s genetics, scientists have been trying to look into our genetic makeup for years. Genome sequencing is what supports the research and helps better determine the complete DNA sequence or makeup of an organism’s genome.
Sequencing is used to identify genetic variants and mutations, which may be associated with health conditions or diseases.Currently, the technology has been used in some medical applications, including prenatal diagnostics and cancer treatment.
Last year, the first draft human genome sequence celebrated its 20th anniversary. This long period of research has resulted in an influx of an unprecedented amount of genomic data. Within the next decade, genomics research will produce between 2 and 40 exabytes of data, thus stalling the progress.
Luckily, complex and gargantuan data analysis is the exact speciality of artificial intelligence and machine learning. By automating genomic data processing, researchers can interpret and act on genomic data in a fraction of the time. Conversely, the traditional gene panel analysis can guzzle up around two weeks to return results.**
To put the speed of AI analysis into perspective, researchers from Stanford University, NVIDIA, and others have built an AI method to do DNA sequencing in 5 hours and 2 minutes. This landed them into the Guinness World Record for the fastest DNA sequencing tool.
Besides lightning-fast data processing, artificial intelligence can help identify variants in the DNA sequence. These variants can be linked to diseases or inherited conditions. Algorithms can also be used to predict how a person will respond to a particular medication based on their genetic makeup. With these insights, researchers gain a better understanding of effective treatment, susceptibility, and disease-causing mutations.
On the same line, real-time nanopore DNA sequencer processing has long remained a challenge for genomics.This unique technology enables direct analysis of long DNA or RNA fragments. Conversely, a slow analysis could lead to decreased survival chances for patients with sepsis and other urgent conditions.
I’m honored to say that our team had a hand in building a trailblazing solution that solved the challenges of real-time nanopore sequencing.
Together with Allen Day, who is a data scientist and senior developer advocate from Google Cloud Platform, we developed a real-time DNA sequence analysis application that detects biocontominants, antibiotic-resistant genes, etc.
Thanks to comprehensive data visualization, medical experts can then make a data-driven decision on a patient treatment plan. The code is available on GitHub.
While the implications for human health have received a lot of attention, genetic sequencing and analysis could also be revolutionary in agricultural and animal husbandry.
Genetic testing is nothing new - but the combination of predictive genetic testing and artificial intelligence is an exciting new development. When it comes to predicting whether or not you may develop a particular disease, predictions are typically very inaccurate.**
However, with the help of artificial intelligence, this may no longer be the case. Thus, researchers from Sutter Health and the Georgia Institute of Technology are using deep learning to analyze electronic health data to avert heart failure. This way, we’re stepping toward a proactive medical approach, instead of leaving it reactionary.
Another example is a team from MIT that developed a deep learning algorithm for preventive melanoma treatment.The algorithm can quickly identify and screen for early-stage melanoma. The latter is responsible for over 70% of skin cancer-related deaths worldwide.**
Genetic screening of newborns is now also amplified with intelligent algorithms. Thus, Children’s National Hospital researchers built a machine learning tool that performs rapid genetic screening.As a result, quick and accurate analysis can accelerate the diagnosis of genetic syndromes in children. Moreover, artificial intelligence can predict outcomes and the risks that stem from curing genetic diseases, based on available data.**
As it’s clear from my post, the synergy between AI and healthcare is present in all medical fields. The application of artificial intelligence for genomic data promises to help researchers better understand the complex genomic profiles of different organisms.
With the help of AI, biologists will be able to take on more ambitious genomics projects and also to more quickly analyze the results of their experiments. The combination of big data and machine learning can help identify patterns in genomic data that otherwise would be difficult for humans to discern. Ultimately, AI and machine learning will accelerate our understanding of our genetic makeup and those of other living organisms.