judftteam / best-of-atomistic-machine-learning Goto Github PK

🏆 A ranked list of awesome atomistic machine learning projects ⚛️🧬💎.

License: Creative Commons Attribution Share Alike 4.0 International

atomistic-machine-learning awesome-list best-of-list computational-chemistry computational-materials-science condensed-matter density-functional-theory drug-discovery electronic-structure interatomic-potentials

best-of-atomistic-machine-learning's People

Contributors

Stargazers

Watchers

Forkers

ml-evs sharadgupta27 dhrubapatra mads-petervc xitanna roshan2004 physroshan flash-jaehyun raolixiang-up neerukulhari sxm13 blaiszik sailfish009 yychuang lhycms zeldery shdchen pythonfz

best-of-atomistic-machine-learning's Issues

Add project: OpenMM ML projects

Community request, moved here from #10.

Project details:

Project Name: OpenMM-ML
Github URL: openmm/openmm-ml
Category: md
Labels: md, mliap
License: MIT
Package Managers: conda:conda-forge/openmm-ml

Project Name: SPICE
Github URL: openmm/spice-dataset
Category: datasets
Labels: datasets, mliap, md
License: MIT
Package Managers: None

Project Name: NNPOps
Github URL: openmm/NNPOps
Category: mliap
Labels: mliap, md, lang-cpp
License: MIT
Package Managers: conda:conda-forge/nnpops

Project Name: openmm-torch
Github URL: openmm/openmm-torch
Category: md
Labels: md, mliap, lang-cpp
License: https://github.com/openmm/openmm-torch#license
Package Managers: conda:conda-forge/openmm-torch

Add project: Open Catalyst datasets

Project details:

Project Name: Open Catalyst datasets
Description: The datasets of the Open Catalyst project, OC20, OC22
URL: https://github.com/Open-Catalyst-Project/ocp/blob/main/DATASET.md
Category: datasets
License: CC-BY-4.0
Resource: true

Additional context:

OC20, OC22

Add project: Materials repositories popular in AML research

Project details:

See below.

Additional context:

This is a tricky subject. Materials repositories. Should they be part of this list, or rather be in a separate "materials informatics" list or similar?

Yes: Other "static" datasets like QM9 are also listed here. Why? Because they are part of atomistic ML (AML) publications as much as the models on which they were trained. Same goes for publications that used materials repos like Materials Project, then.

No. This is a bigger undertaking. There are a lot of materials repositories. Which one should be added? Add only the homepagess as resources, or homepages and APIs as separate entries, or homepages and APIs and associated tools as separate entries?

A balance should be struck here. Adding all kinds of materials informatics and computational materials science tools in this AML list would dilute it. For example, pymatgen and spglib should not be listed here, but in such a dedicated list. So, strike a balance: Only add the minimal set of tools to access materials repo data via browser or via API. Two entries at most per repo as guideline.

List of projects to research

Note. Add only those that actually have a record of being used in published AML research.

Materials Project
AFLOWlib

List of projects to add to the list

Project Name: Materials Project API
Github URL: https://github.com/materialsproject/api
Category: datasets. Or create new category, see below.
License: https://github.com/materialsproject/api/blob/main/LICENSE
Package Managers: pypi:mp-api

Add project: CrystaLLM

Project details:

Project Name: CrystaLLM
GitHub URL: None
URL: https://crystallm.com
Category: community-service
Labels: ?
License: NA
Package Managers: None

Additional context:

https://arxiv.org/abs/2307.04340

Paper with online demo, but no code. But the online demo can be used like a prompt-based crystal structure file generator.

Improve labels

Configuration Change:

Right now, labels have some issues.

Problem 1). Labels with emoji name don't show up in Explanations section.

Custom labels without an image but just a name (emoji or string) are not shown in README section Explanations. That makes some emoji labels confusing. For example, ⚡ for electrostatics can only be guessed at without an "Explanation" entry.

Possible solutions.

Labels with name emoji: replace with an image emoji (an actual image file to be added to config/images). That makes text less copy & paste-able.
Replace all custom labels with name string labels. Then they are all self-explanatory.
Read template docs again and see if I have not misunderstood. Otherwise, create a template issue.

Update project: All where generator failed to gather license

Update details:

Project Name: All where generator failed to gather license, even though it does exist.

Manually add the project's license info in projects.yaml.

Add project: paperswithcode

Project details:

This suggestion is not a single project, but several datasets and benchmark models aggregated via paperswithcode.

For some of the already existing list entries (e.g. QM9, OC20 dataset), maybe replace website with paperswithcode entry.
The paperswithcode tags could be entered as resource in respective category, if any. Collect here first, then assign.
Also add the linked models / packages.

Initial datasets finds.

Initial tags finds.

Add project: EDM

Project details:

Project Name: EDM
Description: E(3) Equivariant Diffusion Model for Molecule Generation in 3D.
Github URL: ehoogeboom/e3_diffusion_for_molecules
Category: generative
Labels: generative
License: MIT
Package Managers: None

Additional context:

Original publication: https://arxiv.org/abs/2203.17003v2

Add project: escnn_jax

Project details:

Project Name: escnn_jax
Github URL: https://github.com/emilemathieu/escnn_jax
Category: representation-learning
License: https://github.com/emilemathieu/escnn_jax/blob/master/LICENSE
Package Managers: pypi:escnn_jax

Additional context:

via https://twitter.com/MathieuEmile/status/1674809578611982338

Add project: mat2vec

Project details:

Project Name: mat2vec
Github URL: https://github.com/materialsintelligence/mat2vec
Category: representations
License: MIT
Package Managers: None

Additional context:

https://www.nature.com/articles/s41586-019-1335-8#data-availability

Add project: DeepCDP

Project details:

Project Name: DeepCDP: Deep learning Charge Density Prediction
Github URL: https://github.com/siddarthachar/deepcdp
Category: ml-dft
License: None
Package Managers: None

Additional context:

https://www.mdpi.com/2079-4991/13/12/1853

Add project: PROPhet

Project details:

Project Name: PROPhet
Github URL: biklooost/PROPhet
Category: ml-dft
Labels: mliap, md, single-paper, lang-cpp
License: GPL-3.0
Package Managers: None

Additional context:

Original publication https://doi.org/10.1038/s41598-017-01251-z

Add project: wfl

Project details:

Project Name: wfl
Description: Workflow is a Python toolkit for building interatomic potential creation and atomistic simulation workflows.
Github URL: libAtoms/workflow
Category: mliap
Labels: workflows, htc
License: None
Package Managers: None

Additional context:

Original publication: https://arxiv.org/abs/2306.11421v1

Use manual description copied from README.md rather than automatic one, that one is too short.

Add project: matsci.org

Project details:

Project Name: matsci.org
URL: https://matsci.org/
Description: "Computational materials science community, online forum."
Category: community-resource

Additional context:

Arguable whether this community resource should be put here, as it is more about general atomistic modeling and simulation, rather than atomistic machine learning. But on the other hand, there is no such dedicated AML community forum, so this is an okay place to put AML question. And AML is a sub- or site field of atomistic modeling will just merge into the general field as time moves on, so why even make a distinction.

Add project: Matter Modeling Stack Exchange - ML

Project details:

Project Name: Matter Modeling Stack Exchange - Machine Learning
Homepage: https://mattermodeling.stackexchange.com/questions/tagged/machine-learning
Category: community

Improve labels II

Configuration Change:

Moved here from previous issue.

Problem 2). Classification by categories is not enough.

Change from categorical (one-dimensional) to tagging (multi-dimensional; here, tags are called 'labels') system. In this system, every list item has labels, and the category is only one of them, the 'main' or 'emphasized' one. Example from issue #10: Package equisolve would gain the labels math, mlp, general-tools, and the category general-tools is the category = emphasized label.

TODO:

Add labels: One for each category (duplicate structure).
Issue #10 add labels for equisolve
Issue #24 add labels for flare

Add project: TensorMol

Project details:

Project Name: TensorMol
Github URL: jparkhill/TensorMol
Category: mliap
Labels: single-paper
License: GPL-3.0
Package Managers: None

Additional context:

Original publication https://doi.org/10.1039/C7SC04934J

Add project: 2DMD dataset

Project details:

Project Name: 2DMD dataset
URL: https://rolos.com/open/2d-materials-point-defects/#physics
Category: datasets
License:

Additional context:

https://doi.org/10.1038/s41699-023-00369-1

Add project: escnn

Project details:

Project Name: E(n)-equivariant Steerable CNNs (escnn)
Github URL: https://github.com/QUVA-Lab/escnn
Category: representation-learning
License: https://github.com/QUVA-Lab/escnn/blob/master/LICENSE
Package Managers: pypi:escnn

Additional context:

More general ML, computer vision. But paper also shows application example to SO(3)-equivariant molecules dataset.

Original publication: https://openreview.net/forum?id=WE4qe9xlnQw.

Add project: ai4material_design

Project details:

Project Name: ai4material_design
Github URL: https://github.com/HSE-LAMBDA/ai4material_design
Category: unsure. representation-learning or representations maybe. maybe need a new category.
License: Apache-2.0
Package Managers: None.

Additional context:

https://doi.org/10.1038/s41524-023-01062-z

Live demo.

https://research.constructor.tech/p/2d-defects-prediction

Add project: CHGNet

Project details:

Project Name: CHGNet
Github URL: CederGroupHub/chgnet
Category: mliap
Labels: md, pre-trained, electrostatics, magnetism, structure-relaxation
License: https://github.com/CederGroupHub/chgnet/blob/main/LICENSE
Package Managers: pypi:chgnet

Additional context:

Add project: WMDGroup's projects

Project details:

AML projects by https://wmd-group.github.io/

Add project: EquiformerV2

Project details:

Project Name: EquiformerV2
Github URL: atomicarchitects/equiformer_v2
Category: rep-learn
License: None
Package Managers: None

Additional context:

The actual model is implemented in the ocp package, see README.

Add project: Atom2Vec

Project details:

Project Name: Atom2Vec
Github URL: https://github.com/idocx/Atom2Vec
Category: representations
License: None
Package Managers: pypi:atom2vec

Additional context:

https://www.pnas.org/doi/10.1073/pnas.1801181115

Note: This seems to be a reimplementation.

Anonymous community feedback

Configuration Change:

Anonymous community feedback 2023-06-13.

Disable license warnings. Which licenses are risky is a matter of opinion.
Add an explanation of the "combined project-quality score".
MD category: Add LAMMPS & all the lammps-xxx repo integrating a given potential with lammps; as well as OpenMM and TorchMD, both usable with ML potentials.

Improvement suggestions to the best-of-generator template.

Allow projects in multiple categories.
- For now, the template only allows one category per project. However, one can implement this also by duplicating a project entry for different categories. That should not require a change to the template. The question is, then, where to draw the line of duplication, in order to stay fair to every package. Ultimately, this shows the deficit of the category (tree-based) vs. label (tag-based) classification schemes. Simplest would be to just do this case-by-case, whenever there is a package developer / community request for it. In that case:
- Put equisolve in multiple categories: math, mlp, general-tools. Developer request.
If the way the project score is calculated raises concern, create an issue with improvement suggestions for .

Add project: MPContribs

Project details:

Project Name: MPContribs
Github URL: https://github.com/materialsproject/MPContribs
Category: community-resource
License: MIT
Package Managers:

Rather use the resource URL.

https://docs.materialsproject.org/services/mpcontribs

Why add this project in this atomistic ML list? Because of "MPContribsML", "Benchmark datasets for Machine Learning".

Add project: Artificial Intelligence for Science (AIRS)

Project details:

Project Name: Artificial Intelligence for Science (AIRS)
Github URL: https://github.com/divelab/AIRS
Category: General Tools or Representation Learning
License: GPL-3.0 license
Package Managers: None

Project Name: AI for Science Map
URL: https://www.air4.science/map
Category: Comunity resource
License: GPL-3.0 license
Description: Interactive visual treemap of the AI4Science research field, including atomistic machine learning, as of summer 2023, including papers, packages, learning resources.

TODO: add missing packages, resources from there here as well.

Project Name: AI for Science Resources
URL: https://github.com/divelab/AIRS/blob/main/Overview/resources.md
Category: Comunity resource, or add to static "similar lists" list in the README footer.
License: GPL-3.0 license
Description: List of resources for AI4Science research, including learning resources.

Update project: Flare

Update details:

Project Name: Flare (https://github.com/mir-group/flare)

Additional context:

The project is listed under General Tools. I think a better fit would be the Active Learning section since a core functionality is building force fields using AL.

Add project: tensorfieldnetworks

Project details:

Project Name: tensorfieldnetworks
Github URL: https://github.com/tensorfieldnetworks/tensorfieldnetworks
Category: representation-learning
License: MIT
Package Managers:

Additional context:

A predecessor to e3nn. Only of historical significance, in that respect.

Add project: SkipAtom

Project details:

Project Name: SkipAtom
Github URL: https://github.com/lantunes/skipatom
Category: representations
License: MIT
Package Managers: pypi:skipatom conda:conda-forge/skipatom

Additional context:

https://www.nature.com/articles/s41524-022-00729-3

Comparison of Word2Vec, Atom2Vec, Mat2Vec, SkipAtom.

https://blog.materialis.ai/distributed-representations-of-atoms.html

Add project: Materials Intelligence's projects

Project details:

AML projects by https://github.com/materialsintelligence

Add project: TorchMD-NET

Community request, moved here from #10.

Project details:

Project Name: TorchMD-NET
Github URL: torchmd/torchmd-net
Category: md
Labels: md, mliap, rep-learn
License: MIT
Package Managers: None

Add project: matbench, MPContribs

Project details:

Project Name: MatBench
Github URL: materialsproject/matbench
Category: community
Labels: community, datasets, benchmarking
License: MIT
Package Managers: pypi:matbench

Project Name: MPContribs
Github URL: materialsproject/MPContribs
Category: datasets
Labels: datasets
License: MIT
Package Managers: pypi:mpcontribs-client

Add project: MEGAN

Project details:

Project Name: MEGAN: Multi Explanation Graph Attention Student
Github URL: aimat-lab/graph_attention_student
Category: xai
Labels: xai, rep-learn
License: MIT
Package Managers: None

Additional context:

Original publication: https://arxiv.org/abs/2305.15961v1

(technically, the original publication is http://arxiv.org/abs/2211.13236, but the two papers link to repos that do not exist anymore or have been updated since. therefore, assume the newer publication is the closest one to current state of code.)

Add project: Visual Graph Datasets

Project details:

Project Name: Visual Graph Datasets
Github URL: aimat-lab/visual_graph_datasets
Category: datasets
Tags: datasets, xai, rep-learn
License: MIT
Package Managers: None

Additional context:

Dataset used in original publication of #69 .

Add project: All from Wasmer's research library

Project details:

Add all missing projects from Wasmer's research library, go to tags (lower left): library, with-code, with-data, database, dataset. For each entry, go to Attachments (upper right) > repository.

Add project: CraTENet

Project details:

Project Name: CraTENet
Github URL: https://github.com/lantunes/CraTENet
Category: unsure. representation-learning, but that is too unspecific. transport properties?
License: MIT
Package Managers: None

Additional context:

https://blog.materialis.ai/predicting-thermoelectric-transport-properties.html

Add project: ML-for-CurieTemp-Predictions

Project details:

Project Name: ML-for-CurieTemp-Predictions
Description: Machine Learning Predictions of High-Curie-Temperature Materials
Github URL: msg-byu/ML-for-CurieTemp-Predictions
Category: rep-eng
Labels: rep-eng, single-paper, magnetism
License: MIT
Package Managers: None

Additional context:

Original publication: http://arxiv.org/abs/2307.06879.

Add project: cdk

Project details:

Project Name: cdk
Github URL: cdk/cdk
Category: rep-eng
Labels: cheminformatics, lang-java
License: LGPL-2.1
Package Managers: pkg:maven/org.openscience.cdk/cdk-bundle

Additional context:

Add project: MateriApps

Project details:

Project Name: MateriApps
URL: https://ma.issp.u-tokyo.ac.jp/en/
Category: Community Resources, or "Other lists" (footer.md)
License: NA?

Additional context:

What’s MateriApps

A Portal Site of Materials Science Simulation for Computational Materials Science Researchers, Theoreticians, Experimentalists, and Computer Scientists

Add project: Citrine Informatics ERD projects

Project details:

Project Name: GlassPy
Github URL: drcassar/glasspy

Category: rep-eng
Labels: rep-eng glass, composition, materials
License: GPL-3.0
Package Managers: pypi:glasspy

Are compositional feature -- based models atomistic models? In this case, I decide for "Yes".

Project Name: SciGlass
Github URL: drcassar/SciGlass

Category: datasets
Labels: datasets, glass
License: MIT

Project Name: sl_discovery
Github URL: CitrineInformatics-ERD-public/sl_discovery
Category: materials-discovery
License: Apache-2.0
Package Managers: None
Labels: materials-discovery, single-paper

Project Name: Linear vs blackbox
Github URL: CitrineInformatics-ERD-public/linear-vs-blackbox
Category: xai
Labels: xai, single-paper, rep-eng
License: None

Project Name: closed-loop-acceleration-benchmarks
Github URL: aced-differentiate/closed-loop-acceleration-benchmarks
Category: materials-discovery
Labels: materials-discovery, active-learning, single-paper
License: MIT
Package Managers:

Project Name: EquivariantOperators.jl
Github URL: aced-differentiate/EquivariantOperators.jl
Category: math
Labels: math, symmetrized, lang-julia
License: MIT

Project Name: Closed-loop acceleration benchmarks
Github URL: aced-differentiate/closed-loop-acceleration-benchmarks
Category: materials-discovery
Labels: materials-discovery, active-learning, single-paper
License: MIT
Package Managers:

Project Name: Citrination Datasets
URL: https://citrination.com/datasets
Category: datasets
Labels: datasets
License: MIT
Package Managers: None.

Add project: mlp (Proper orthogonal descriptors, POD)

Project details:

Project Name: mlp
Description: "Proper orthogonal descriptors for efficient and accurate interatomic potentials. Paper."
Github URL: https://github.com/cesmix-mit/mlp
Category: mlp
Labels: lang-julia
License: None
Package Managers: None

Additional context:

POD descriptor benchmark, as referenced in the corresponding paper https://doi.org/10.1016/j.jcp.2023.112030.

Original publication not linked in README as of 2023-12-03. Repo description not descriptive. Replace with custom description: Paper title plus link to paper.

Add publication metrics

Configuration Change:

See correspondent upstream issue.

Add project: Materials Project Charge Densities & mp-pyrho

Project details:

Project Name: Materials Project Charge Densities & mp-pyrho
Github URL: https://github.com/materialsproject/pyrho
Category: data-structures for mp-pyrho, datasets for MP Charge Densities
License: https://github.com/materialsproject/pyrho/blob/main/LICENSE
Package Managers: pypi:mp-pyrho

Additional context:

URL for the MP CD datasets project entry:

https://next-gen.materialsproject.org/ml/charge_densities

The CDs are really part of MP API now. But still worthwhile to emphasize with a separate entry that they exist in the first place.

Add project: DIVE Lab's projects

Additional context:

Projects by DIVE Lab.

Add project: Add all LAMMPS ML potentials

Community request, moved here from #10.

MD category: Add LAMMPS & all the lammps-xxx repo integrating a given potential with lammps; as well as OpenMM and TorchMD, both usable with ML potentials.

Subtasks OpenMM, TorchMD moved to issues #77, #78.