Thanks to artificial intelligence (AI) machine learning and applied physics, milestones in biology have been achieved. A new peer-reviewed study by researchers at Harvard’s John A. Paulson School of Engineering and Applied Sciences (SEAS) and Northwestern University used AI and physics to engineer the unstable proteins that make up the dark proteome.
This shows how AI can accelerate research, especially in the fields of biology, synthetic biology, and personalized medicine.
“The design of folded proteins has advanced significantly in recent years,” co-authors Krishna Shrinivas, Michael P. Brenner, and Ryan K. Krueger wrote. “However, many proteins and protein regions are inherently disordered and lack stable folds. In other words, the sequences of intrinsically disordered proteins (IDPs) encode vast ensembles of spatial structures that specify their biological function.”
dark proteome
Similar to physics and modern cosmology, where there is dark matter, there is a dark proteome in biology. The exact composition of dark matter is unknown, although a 2022 study by Yuan et al. estimated that it makes up 27% of the universe’s total mass-energy density and 85% of the matter in the universe. Published in physical report. The dark proteome may contain a potential treasure trove of undiscovered therapeutic targets for drugs to fight disease. The dark proteome includes proteins with unknown structures, less studied proteins, undetected proteins, proteins derived from non-canonical genes that deviate from canonical genes, and proteins that are inherently disordered.
Intrinsically disordered proteins do not have a well-defined structure. The lack of a stable structure makes it nearly impossible to study IDPs with traditional structural biology tools such as X-ray crystallography and cryo-electron microscopy.
The human proteome is the complete set of proteins expressed in the human body, consisting of approximately 20,000 proteins, according to an estimate published in 2006. natural chemical biology According to Eibersold et al. According to Bind Research, an estimated 65% of the human proteome is composed of structured proteins, and the remaining 35% are inherently disordered proteins.
Why study proteins?
Understanding proteins is essential to many fields such as neurodegenerative disease research, drug discovery, drug design, biotechnology, synthetic biology, protein engineering, personalized medicine, and medical research. Many diseases and neurodegenerative diseases are caused by protein misfolding and protein aggregation. Decreased protein stability or disruption of protein folding can lead to the formation of protein aggregates, which can result in the accumulation of unfolded or misfolded proteins.
For example, the accumulation of misfolded tau protein causes neurofibrillary tangles called tau changes, which are one of the signs of Alzheimer’s disease. Alzheimer’s disease research By Bright Focus Foundation. Misfolded alpha-synuclein protein is a biomarker for Parkinson’s disease, according to a study published in lancet neurology It will be led by Professor Claudio Soto of the University of Texas McGovern Medical School in Houston. According to a study published in , one of the signs of Huntington’s disease is misfolding of the huntingtin protein. brain research By Jean et al. Al. Protein misfolding leading to protein aggregation is seen in amyotrophic lateral sclerosis (ALS).
Beyond Nobel Prize-winning AI
The successful application of AI deep learning to predict the 3D structure of proteins began in 2018, with Google DeepMind’s AlphaFold leading the way by achieving the highest accuracy among 13.th Critical assessment of competition in protein structure prediction (CASP). AlphaFold is an AI deep learning model that predicts the three-dimensional structure of proteins from amino acid sequences. Two years later, in 2020, AlphaFold 2 achieved the highest accuracy in the CASP14 competition and was recognized by CASP organizers for solving a 50-year grand challenge in biology: protein folding.
The 2024 Nobel Prize in Chemistry was awarded to David Baker for computational protein design and to Google DeepMind AI pioneers Demis Hassabis and John Jumper for protein structure prediction. This was a pioneering study to predict single 3D structures of stable proteins.
However, designing inherently disordered proteins using AI deep learning is inherently difficult due to the nature of machine learning. Deep neural networks “learn” features from large amounts of training data, rather than through hard-coded explicit computer programming. While the protein datasets used to train AlphaFold may have included IDP data, the majority contained information about structured proteins captured by standard biological tools such as X-ray crystallography. Furthermore, according to the AlphaFold protein structure database, AlphaFold is more likely to make unreliable predictions regarding intrinsically disordered regions (IDRs), which are protein segments that do not have a stable 3D structure.
Even though AlphaFold 2 was trained primarily on stable proteins, it still encodes “important stability information.” physical review letter Written by McBride and Trusty. This opens the door to understanding unstable proteins.
innovative approach
A seemingly obvious way to conduct this new research is to identify a database consisting primarily of inherently disordered proteins and train a deep learning algorithm to predict the structure of the proteins. But researchers at Harvard University and Northwestern University chose to take the road less traveled in their approach. What makes this new study unique is that the scientists used gradient-based optimization to create an AI model that applied fundamental physical laws in realistic molecular dynamics simulations.
“The combination of physics-based approaches and recent advances in differentiable programming may enable the computational design and engineering of a wide variety of biomolecules and their functions,” the study authors reported.
This pioneering research uses applied physics and AI to shed light on one of the key components of the dark proteome, potentially leading to new therapeutic targets and novel treatments to fight a wide range of diseases in the future.
Copyright © 2025 Kami Rosso All rights reserved.
