In the past two decades, there have been unprecedented advancements in many frontiers of technology. But none have made more buzz than Artificial Intelligence (AI). Due to its nature, AI has been applied almost everywhere, be it academia, research or industry. As a result, its impact is visible on various domains and sub-domains. Here, we will discuss the impact of AI, specifically in biology.
What is AI
In animal behavior studies or neuroscience, sometimes we are interested in knowing whether the subject animal is intelligent or not. So, different experiments are designed to test their skills to understand the nature and the extent of their intelligence. But, what about machines like computers?
During my childhood, we were taught that computers couldn’t think for themselves. For any behavior that we could observe, rules and scenarios were programmed explicitly. That meant computers can only follow instructions (and they were exceptional in doing that). These implied computers are not intelligent—however, AI suggests just the opposite.
As per the famous ‘Turing test,’ developed by British mathematician Alan Turing, true AI is when a human is deceived (can’t distinguish between) by a computer or any machine as a fellow human being under specific conditions. In real-world applications, using AI, people try to mimic or sometimes surpass human intelligence and decision-making.
Rise of AI
The theoretical aspect of AI has started with Alan Turing in the 1940s. But the practical application side of it remained stagnant for most of the next 50 years. The reasons for it are the same as that of the rise of AI in the last two decades- data and computational power.
The core of intelligence is learning, and for learning, you need a huge amount of data. In recent years, digitization had become a central theme; millions of Gigabytes (GB) of data are created every day. This massive amount of data has been an immense help in creating AI models and training for specific tasks. Another important necessity for AI is the computational power to utilize the enormous amount of data. In addition, we have access to powerful hardware with unparallel processing speeds. With these two things, we have seen such a spread of AI practically everywhere.
Computers are capable of ingesting and processing large amounts of data, and now with the use of AI, they can be used to look for complex patterns.
AI in Biology
Biology is a vast subject and thus has innumerable possibilities for AI application. Here, we will discuss a few major topics and use cases where the use of AI has been a boon. This will also hopefully demonstrate for interested readers other plausible applications of AI.
Imaging and Computer-aided Diagnosis
One of the major use cases of AI application on biology can be seen in healthcare. Multiple diagnoses rely on different imaging techniques like MRI, CT scan, PET scan, X-ray, etc. Trained radiologists diagnose such conditions (or absence of it) based on the image data and their expertise. Now with the use of AI, models are trained with a huge number of annotated data (where the corresponding ailment is already known). Once appropriately trained with enough diversity of data, the model can predict the correct diagnosis at par (or even better) with a human specialist. One popular example is detecting pneumonia from lungs X-ray or lungs CT scan images. This sort of automation has been especially helpful in spreading health benefits in developing nations where the doctor-to-patient ratio is significantly less. Based on data availability, the classification can be simple binary (like, a yes/ no question or presence/ absence of tumors) or much complex multi-class classification (detecting and cataloging different cell morphologies from a microscopic image, which usually require a human observer).
Apart from diagnosis and health-related imaging scans, AI is increasingly used in other imaging-based research studies. Computers are capable of ingesting and processing large amounts of data, and now with the use of AI, they can be used to look for complex patterns. For example, few research labs are trying to look for the correlation between the progression of Alzheimer’s disease and fMRI (functional-MRI) data for certain behavioral tasks in comparison to resting state.
AI in Protein folding
Another very recent progress of AI application has been seen on the protein-folding problem. It is one of the most challenging and complex problems in structural biology for the last 50 years or so. Proteins are responsible for all of what happens inside of our bodies. The wide variety of functions proteins execute depends on their 3-dimensional shape. Based on the amino acid sequence (Amino acids are the building blocks of proteins), different forces and interactions come into play to form the particular 3-d shape. Due to the large number of factors involved, it remained a challenging problem to predict a protein shape or how it folds from a given amino acid sequence. Experimental methods like Cryo-electron microscopy, NMR, X-ray Crystallography, etc., are used to deduce a 3-D structure of a protein. These are hugely time- and cost-consuming methods.
In recent years, AI-based techniques did exceptionally well to solve the protein folding problem. For example, in 2020, a model named ‘AlphaFold,’ developed by Deep mind, did so well in CASP14 (a biennial competition for solving protein folding problem) that a large portion of the scientific community think that the protein-folding problem might have been solved. This is one such example where AI capabilities surpassed human achievement.
AI in Neuroscience
The effect of AI in neuroscience has been bi-directional. Both have benefitted from the research progress in the other. On the one hand, like other biology topics, AI has been an indispensable tool in neuroscience research. On the other hand, understanding the human brain has led to better and complex AI models- capable of capturing complex data from the real world.
I have previously talked about imaging techniques. A large portion of neuroscience makes use of such imaging techniques. AI-based models are great tools for image analysis and finding complex patterns. AI models also work as a virtual brain to study how some of the higher-order brain functions might work. Similarly, Deep learning models, a specific type of AI model, are inspired by human brain structures. Concepts like attention, the hierarchy of order of functions, etc., are taken from neuroscience and successfully applied to AI models.
AI in Agriculture
To demonstrate the widespread applicability of AI, below are some examples of AI in the agricultural space as well. One of the major use cases in this area is to detect plant diseases based on images taken from devices like- mobile, drones, IP cameras, etc. The use of AI models has also enabled robotic process automation in farming in developed nations.
AI in Sequence data
The last couple of decades have seen the collection of colossal sequencing data. Both genome sequencing data and protein sequence data are meticulously collected and annotated by different research groups. These data are used in junction with AI models for various analytical and predictive purposes. For example, predicting immunogenicity of a particular antigen based on the provided sequence.
Conclusion
For every use case or type of application mentioned, tons have been missed. AI has implications from ecology to epidemiology, molecular biology to system biology, and many more. With every day, AI has become more user-friendly, and more and more research groups are adapting AI models for various purposes. Hopefully, this article has given a glimpse of the diverse set of applications that AI can be used for. A future biologist might use AI for the next big breakthrough, and as such general awareness regarding AI must increase as well.
References and Further reading
- R. F. Service, “‘The game has changed.’ AI triumphs at protein folding,” Science (80-.)., vol. 370, no. 6521, 2020, doi: 10.1126/science.370.6521.1144.
- K. A. Dill, S. B. Ozkan, M. S. Shell, and T. R. Weikl, “The protein folding problem,” Annu. Rev. Biophys., vol. 37, pp. 289–316, 2008, doi: 10.1146/annurev.biophys.37.092707.153558.
- Webb, S., 2018. Deep learning for biology. Nature, 554(7690), pp.555-558.
Author Information
Sourav Mukherjee did his BS-MS (Integrated dual degree) from the Indian Institute of Science Education and Research, Pune. His interests during this time were Number Theory, Mathematical biology, Mathematical modeling, and Neuroscience. In 2017, he spent a year at the Indian Institute of Science, Bangalore, studying selective attention in humans. Afterward, he spent almost a year in Indian Statistical Institute, Kolkata, studying the correlation between attentional behavior and emotion recognition in humans. From 2019 onwards, Sourav holds the position of Lead Data Scientist in Bayshore Intelligence Solutions (India) Pvt. Ltd.
Graphics by: Jyotirekha Das
Edited by: Manveen K Sethi