$ aixsci models
Named AI models in science
Models that studies use by name, from protein structure predictors to galaxy classifiers: what each one is, and what researchers actually did with it.
transformer
AlphaFold
AlphaFold is a computer system that reads a protein's chain of chemical building blocks and predicts the shape that chain folds into, without any laboratory experiment.
gradient-boosted trees
XGBoost
XGBoost is a piece of free software that builds a prediction from hundreds of small decision trees, each one added to correct the mistakes of the trees before it.
protein language model
ESM-2
ESM-2 is a model trained on millions of natural protein sequences, which turns a chain of amino acids into lists of numbers that other programs can learn from.
protein language model
ProtT5
ProtT5 is a computer model trained on vast numbers of protein sequences, which turns a protein's chain of amino acids into a list of numbers that other programs can learn from.
transformer
ColabFold
ColabFold is a free, quicker way to run AlphaFold, the program that predicts a protein's three-dimensional shape from the string of chemical letters that make it up.
diffusion model · transformer
AlphaFold 3
AlphaFold 3 is a computer program that takes the chemical recipe of a molecule and predicts the shape it folds into, including proteins sitting alongside DNA, RNA or small drug-like molecules.
protein language model
ESMFold
ESMFold is a computer tool that guesses the three-dimensional shape of a protein from its chain of building blocks alone, without first hunting for related proteins to compare it with.
graph neural network
ProteinMPNN
ProteinMPNN is a neural network that reads the three-dimensional shape of a protein's backbone and suggests chains of amino acids that would fold into that shape.
large language model
ChatGPT and GPT models
ChatGPT and the GPT models behind it are general-purpose text systems made by OpenAI, trained to continue and respond to writing, and now borrowed by scientists as all-purpose assistants.
gradient-boosted trees
LightGBM
LightGBM is a free software tool that learns to predict a number or a label from a table of data, by building many small decision trees, each one correcting the mistakes of those before it.
convolutional neural network
crYOLO
crYOLO is a neural network that scans noisy electron microscope images and marks where each copy of a frozen molecule sits, doing by itself a job researchers once did by hand.
graph neural network
MACE
MACE is a family of machine-learned models that predict the forces between atoms, letting a computer simulate matter far faster than solving the underlying physics equations directly.
convolutional neural network
BTSbot
BTSbot is a neural network that reads the stream of alerts from a sky survey and picks out the bright exploding stars worth pointing another telescope at.
gradient-boosted trees
CatBoost
CatBoost is a free software library that builds a prediction from many small decision trees, each one added to correct the mistakes of the ones before it, and it copes well with labelled categories.
graph neural network
M3GNet
M3GNet is a trained computer model that guesses the energy of a group of atoms, and the forces pulling on each one, for materials made from almost any element in the periodic table.
multilayer perceptron
PONDR
PONDR is a family of computer programs that read a protein's sequence and guess which parts of it stay floppy and shapeless rather than folding into a fixed form.
protein language model
ProtBert
ProtBert is a computer model that learns the patterns of protein sequences by reading millions of them, in the way a text model learns language, and turns each protein into numbers.
diffusion model · transformer
Boltz
Boltz is a family of freely released computer models that work out the three-dimensional shapes of proteins and the ways they stick together, and can also propose new ones.
convolutional neural network
braai
Braai is a neural network that looks at images from a sky survey and decides which flickers of light are real cosmic events and which are just faults in the picture.
graph neural network
CHGNet
CHGNet is a machine learning model that predicts the energy and forces of a group of atoms from the way they are arranged, while also tracking the charge on each atom.
protein language model
ESM-1v
ESM-1v is a computer model that has read millions of protein sequences and can guess, without any experiment, whether swapping one building block in a protein will harm how it works.
protein language model
ESM-IF1
ESM-IF1 is a computer model that works backwards from a protein's three-dimensional shape and proposes strings of amino acids that could fold into it.
model
GEMME
A way of guessing whether a change to a protein's building blocks will harm it, by reading how that protein has drifted and held firm across the history of living things.
model
IUPred
IUPred is a computer tool that reads a protein's sequence and marks which stretches are likely to stay floppy and shapeless rather than fold into a fixed form.
graph neural network
ModelAngelo
ModelAngelo is a program that reads a three-dimensional map of a protein made by electron microscopy and works out, atom by atom, which protein chain fits inside it.
convolutional neural network
Zoobot
Zoobot is a family of image-recognition models trained on the shape judgements that volunteers made about galaxies, and then reused as a ready-made starting point for new astronomical image tasks.
diffusion model · transformer
Chai-1
Chai-1 is a computer model that takes the sequence of a protein and works out the shape it folds into, together with the shapes of the molecules that stick to it.
diffusion model
RFdiffusion
RFdiffusion is a computer model that designs new protein shapes from scratch, starting from random static and refining it, step by step, into a plausible folded structure.
transformer
RoseTTAFold
RoseTTAFold is a computer program that works out the three-dimensional shape of a protein from the sequence of chemical units it is built from, without doing any laboratory experiment.