Fréchet ChemNet Distance: A Metric for Generative Models for Molecules in Drug Discovery

By Kristina Preuer, Philipp Renz, Thomas Unterthiner, Sepp Hochreiter, Günter Klambauer. Journal of Chemical Information and Modeling, 2018

454 citations recorded by the source corpus.

The new wave of successful generative models in machine learning has increased the interest in deep learning driven de novo drug design. However, method comparison is difficult because of various flaws of the currently employed evaluation metrics. We propose an evaluation metric for generative models called Fréchet ChemNet distance (FCD). The advantage of the FCD over previous metrics is that it can detect whether generated molecules are diverse and have similar chemical and biological properties as real molecules.

Subfields: Computational Theory and Mathematics, Materials Chemistry, Molecular Biology

View Paper: https://www.semanticscholar.org/paper/7d878fe31b9b57f75071586d83cdec2e8b81e039

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Theory and Mathematics

In Materials Chemistry

In Molecular Biology

Citations in the corpus

Leveraging molecular structure and bioactivity with chemical language models for de novo drug design (2023, methodology): “This improvement occurs in terms of (i) the quality of the SMILES strings generated, as re ﬂ ected by their validity (gram-matically valid SMILES strings), uniqueness (nonrepetitive molecules), and novelty (molecules not present in the pretraining and ﬁ ne-tuning data) and (ii) the similarity of the”
Artificial intelligence in drug discovery: recent advances and future perspectives (2021, methodology): “…than predictive methods, some of the commonly used metrics are: (i) validity of the generated molecular representations and novelty of the corresponding molecules, (ii) similarity to known compounds in terms of chemical and biological properties [157], and (iii) scaffold and fragment diversity.”
Towards Accurate Generative Models of Video: A New Metric & Challenges (2018, methodology): “FVD builds on the principles underlying Fréchet Inception Distance (FID; [14]), which has been successfully applied to images, and was later adapted to other fields [31].”
Translation between Molecules and Natural Language (2022, methodology): “Preuer et al. (2018) propose Fréchet ChemNet Distance (FCD), which is inspired by the Fréchet Inception Distance (FID) (Heusel et al., 2017).”
Improving Molecular Design by Stochastic Iterative Target Augmentation (2020, methodology): “By contrast, we observe that the REINVENT baseline heavily degrades performance on the FCD metric compared to the baseline on both tasks.”
Randomized SMILES strings improve the quality of molecular generative models (2019, methodology): “Nevertheless, the Fréchet ChemNet Distance (FCD) [36] between the validation set and a sampled set of 75,000 SMILES is lower on the canonical SMILES model.”
Therapeutics Data Commons: Machine Learning Datasets and Tasks for Drug Discovery and Development (2021, background): “[112] Kristina Preuer, Philipp Renz, Thomas Unterthiner, Sepp Hochreiter, and Gunter Klambauer.”
Score-based Generative Modeling of Graphs via the System of Stochastic Differential Equations (2022, methodology): “Moreover, GDSS significantly outperforms all the baselines in NSPDK MMD and most of the baselines in FCD, showing that the generated molecules of GDSS lie close to the data distribution both in the space of graphs and the chemical space.”
Leveraging large language models for predictive chemistry (2024, unknown): “To quantitatively capture the similarity of the distribution of the generated molecules to the ones from the training set, we compute the Fréchet ChemNet distance 64 , which quantifies both diversity and distribution match 51 and goes through a minimum at intermediate temperatures.”