23

The short answer is yes. Machine learning, data mining, AI and other techniques are highly useful in chemistry. I completely agree with Fred's answer that lots of machine learning, expert systems and statistical analysis in chemistry goes back a long time. This is particularly true in analytical chemistry - match a mass spec or NMR or IR against a library ...


14

There is one Python package that I know: AMP (Atomistic Machine-learning Package). It is based on ASE (Atomic Simulation Environment), a homogeneous interface to a lot of computational chemistry packages. Basically, after you obtain geometry trajectories through ASE (using any computational chemistry package you like), an approximation to the potential ...


14

Yes. It's not a new thing at all in chemistry; you'll find papers on chemical AI applications going back four decades or more. We have massive amounts of data that must be processed quickly. For example, a diode array detector in HPLC can collect large numbers of spectra per minute at hundreds of wavelengths, and we need to use that data to distinguish and ...


14

This sounds like you were exploring work at least related to the work by the Lilienfeld group equally hosting a dedicated site here about data sets already used in their earlier and ongoing exploration of chemical space, programs used to work with the data, and publications. To go considerably higher in molecule count than QM9, you could either go for GDB-...


11

The ISOL24 database (http://www.thch.uni-bonn.de/tc.old/downloads/GMTKN/GMTKN55/ISOL24.html) contains molecules with up to 81 atoms! The other answer says that there's a database called "OE" with molecules that have up to 174 atoms, but it is "not yet publicly available".


8

Are you asking how to take code developed to find the MFE (minimum free energy) structure of a single protein, and modify it to instead determine the structure of a two-protein complex? Or are you asking if you can use this code, as is, to fold a concatenated combination of the two proteins, and in turn use this result as a surrogate for the predicted ...


7

Yes, you're exactly right - multiple papers in chemistry ML drop the units. There are even comparisons (usually by statistics, ML or comp. sci. researchers) where models are compared by "averaging" errors down a column like that. Of course that's meaningless, since you can't average electron volts or Hartree (energies), Debye (dipole moments), and volume (...


6

I agree that there seems to be a shift towards using graph representations over SMILES strings. I personally think this is a good thing, and I'll try to explain why, but even if there's nothing inherently better about graph representations of molecules, there is a very practical reason people are moving towards graph representations. So, first of all, as ...


6

TLDR: Defining the similarity between two molecules is an open field of research, but is fundamentally a subjective question which depends on your context. If you are interested in the similarity of molecular structures, then a graph-based measure might be a logical choice. If you're interested in similarities of reactivity, then a measure incorporating the ...


5

Beyond other answers, I'd suggest the original PubChemQC project, which offers ~3 million molecules from PubChem optimized using DFT (B3LYP/6-31G*). Molecules include a wide variety of elements as long as the molecular mass is less than 500 Da. (Roughly speaking that should still handle ~38 carbon atoms.) "PubChemQC Project: A Large-Scale First-Principles ...


5

I know I'm a little late to this party, but in the last few years there have been some potentially very important developments in the application of machine learning to chemistry. Both of which apply to molecular dynamics. First, one potentially obvious observation is that when performing a molecular dynamics simulation, each of the time steps is highly ...


5

In terms of how they got the relation for the diagonal elements, I believe it is relatively straightforward. Given a list of atomic energies (energy of atom relative to separated nucleus and electrons), one can try to fit these values to a function of nuclear charge. If you plot a few values, it suggests trying an exponential fit (I don't know why all the ...


5

DeepChem's project web page indicates further documentation and access to the source files used. Just click the «fork me on github» sign in the top right hand corner leads you to this page. From the notes there, it becomes obvious that many tasks the program suite handles are managed with Python which you should install in first place (e.g., Python's ...


4

What you have provided is a definition of all the terms involved. This is fine, but there remains the (rather interesting) question of exactly how this is defined for a molecule in silico, especially because your question asks specifically about the GDB-13 database. Of course, one could evaluate "synthetic accessibility" by trying to make it in the lab: if ...


4

It depends on how you code your molecular graphs The idea of a 'connection table' or valence model for molecules, and thus molecular graphs is embedded in chemical thinking. Let's take your four points: It's possible to design connection tables that supports a variety of interactions. For example, zero order bonds can encode coordination bonds, ...


3

It is true that a high order polynomial can fit any training set. But that is not a strength - an unfalsifiable model overfits. In particular, a polynomial of order n is only likely to be predictive if the true function is n times differentiable. Since chemical space is discrete, and for many purposes some molecules are special cases, polynomial models are a ...


3

For n data points there exists an n-1 order polynomial that perfectly fits the data. Therefore there is no basis for "a neural network or whatever" being better. Furthermore, it simply isn't true that the Peng-Robinson equation "has no underlying physical meaning". The Peng-Robinson equation (like Van der Waals) recognizes that atoms/molecules occupy ...


3

I don't know about that particular GitHub archive, although I'll point out that the smiles.csv contains several kinds of SMILES - the final column containing the correct SMILES. IMHO, this is an unfortunate result of people mis-using SMILES. As you note, it's perfectly possible to take a chiral molecular structure and output a SMILES with no stereochemical ...


2

After spending the afternoon googling around I've written a simple answer to my own question (essentially a basic definition of terms). For a more complete answer see the accepted answer by orthocresol♦: Synthetic Accessibility - Refers to ease of synthesis. I.e how difficult a compound is to make (synthesize) in a lab. https://jcheminf.biomedcentral.com/...


2

If you're interested in metal complexes, there's a free software, Python-based tool called molSimplify. From their site: Geometry optimization with density functional theory (DFT), a general procedure to obtain the ground state structures of a complex, is computationally demanding in terms of time and can also easily fail. Two main failure modes are ...


2

As several comments have mentioned, interpreting a machine learning method can often be difficult. Outside the question of machine learning, I think your question centers around "what would I want to know to roughly predict HOMO / LUMO energies?" Unfortunately QM9 isn't very diverse chemically, but if we think from a general qualitative molecular orbital ...


2

It's not an all-or-nothing situation, where either we can reliably prediction reaction products, or we have not tried to. For instance, drug companies, for many years, have been making extensive use of increasingly powerful, sophisticated, and refined in silico predictive models to choose likely candidates for in vitro testing. And I would be surprised ...


Only top voted, non community-wiki answers of a minimum length are eligible