fields/structural-biology/what the AI was for
Classification in structural biology and protein science
How researchers in structural biology and protein science use artificial intelligence to sort proteins and the parts of proteins into categories, such as ordered or disordered, binding or not, working or broken.
How it is used
Structural biology studies the shapes that proteins and other large molecules take, and protein science asks how those shapes relate to what a protein does. Classification here means putting an object into one of a set of known boxes. The objects vary in size. Sometimes it is a whole protein, which may be labelled as belonging to a family of interest or not. Sometimes it is a short stretch of the chain, marked as a candidate binding site. Sometimes it is a single building block, one amino acid, scored as lying in a rigid part of the chain or a floppy one. In the examples, the input is usually the sequence alone, the string of letters naming each amino acid in order.
Models of several kinds appear. Some are hand-written rules that look for stretches rich in particular amino acids. Some convert a sequence into a list of numbers, measures of charge, oiliness, stability and the tendency to form particular local shapes, and feed those into a trained classifier. Some use a protein language model, a network trained on many sequences to predict missing letters, which yields a numeric summary of each position that other tools can read. Structure predictors also contribute, supplying features such as how buried a residue is or how many neighbours it touches.
Why researchers use it
Laboratory work that settles these questions is slow. Determining a structure, or measuring whether a protein binds its target, takes effort and material, and the number of known sequences far exceeds the number of solved structures. Classification from sequence stands in for that work as a filter: it narrows a large list to the cases worth testing, or annotates a protein nobody will ever crystallise.
It also stands in for human judgement applied one case at a time. An expert can look at a sequence and guess which region is disordered or which segment might grip a strand of DNA, but cannot do so for a whole genome. The examples show classification wired into larger pipelines: trimming floppy ends before designing an assay, deciding which variants are active enough to be worth measuring further, or defining regions where several independent predictors agree.
In the record
Limits
These models learn from the proteins already labelled, and those lean towards well-studied families, so a sequence unlike anything in the training set may be sorted confidently and wrongly. A category label hides a great deal: calling a region disordered says nothing about what it does, and different predictors disagree often enough that consensus procedures are needed. Features taken from predicted structures inherit the prediction's errors. And a sequence-based label is a hypothesis about a molecule in isolation, whereas real proteins behave differently depending on partners, chemical modifications and surroundings.
Written by AIxSci: the background from general knowledge, the examples from checked records.
Related
Every article
aix-00273
Deep learning model flags bacterial proteins likely to drive infection
aix-00157
Microscopy and a trained classifier map which E. coli proteins clump together
aix-00156
Protein language model and two classifiers predict where ATP binds on proteins
aix-00151
Classifier sorts true from false AlphaFold predictions of human protein pairs
aix-00162
Software splits coiled-coil proteins into short windows for AlphaFold to model
aix-00166
Neural network assigns protein domains to structural families from sequence alone
aix-00168
Deep learning reads cryo-EM maps alongside AlphaFold3 models to build protein structures
aix-00169
AlphaFold's confidence scores used to predict which disordered protein segments bind LC8
aix-00172
Machine learning tested against three definitions of protein shape-shifting
aix-00145
Repeating-coil proteins designed from random sequences using AlphaFold2 in an evolution loop
aix-00143
Software models of two cancer proteins used to rank mutations by likely effect
aix-00140
Model labels which protein residues are active sites and what job they do
aix-00138
AI-labelled protein pairs used to test how well function predictors handle unknown proteins
aix-00179
Neural network predicts protein pair binding strength from simplified molecular models
aix-00181
Protein language model embeddings used to predict succinylation sites in proteins
aix-00184
Machine learning scores short protein stretches that mark proteins for destruction
aix-00185
Protein language model trained to spot lactylated lysines in rice proteins
aix-00188
Machine learning sorts antibody heavy chains by the germ they target
aix-00128
Phage proteins drawn as pictures, then sorted by image-recognition networks
aix-00121
Machine learning sorts mitochondrial proteins into three compartments from sequence alone
aix-00116
Neural network reads protein surfaces to predict DNA and RNA binding
aix-00115
A shared test set for predicting which enzyme carries out a reaction
aix-00112
Protein language model embeddings tested against classical descriptors for spotting resistance genes
aix-00110
Deep learning trained on robotic peptide mapping predicts where antibody proteins degrade
aix-00107
Machine learning sorts plant protein sequences into disease-resistance proteins or not
aix-00081
Language models trained on protein sequences used to predict sugar-attachment sites
aix-00074
Software predictions map floppy stretches of touch-sensing PIEZO channels
aix-00072
Attention patterns inside a protein language model used to cut sequences into reusable units
aix-00050
Machine learning tool flags protein helices likely to bind DNA or RNA
aix-00024
Machine learning picked which engineered enzymes to build for fatty alcohol production
aix-00009
Gaussian process models pick light-sensitive ion channels that reach mammalian cell membranes
aix-00004
Robot lab and learning agent redesign an enzyme to survive higher temperatures
aix-00001