Exploring Graph Traversal Algorithms in Graph-Based Molecular Generation

Mercado, Rocío; Bjerrum, Esben Jannik; Engkvist, Ola

doi:10.1021/acs.jcim.1c00777

Cited by 9 publications

(4 citation statements)

References 23 publications

(47 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This has been achieved by representing every molecular structure as a graph, where the atoms are the nodes, and the bonds are the edges. Using the depth-first search (DFS) algorithm [ 29 ], the innovative idea is to identify the corresponding atoms of the two structures by following the same path from an atom selected as the root to the leaves, matching the same atoms in the two molecular structures between which the RMSD is calculated. The match between two atoms, one in the reference structure and the other in the predicted pose, is based on: ( i ) the atom type, ( ii ) the covalently bound atoms, and ( iii ) the order of such bonds.…”

Section: Resultsmentioning

confidence: 99%

PacDOCK: A Web Server for Positional Distance-Based and Interaction-Based Analysis of Docking Results

Carbone

Ghidini

Romano³

et al. 2022

Molecules

View full text Add to dashboard Cite

Molecular docking is a key method for structure-based drug design used to predict the conformations assumed by small drug-like ligands when bound to their target. However, the evaluation of molecular docking studies can be hampered by the lack of a free and easy to use platform for the complete analysis of results obtained by the principal docking programs. To this aim, we developed PacDOCK, a freely available and user-friendly web server that comprises a collection of tools for positional distance-based and interaction-based analysis of docking results, which can be provided in several file formats. PacDOCK allows a complete analysis of molecular docking results through root mean square deviation (RMSD) calculation, molecular visualization, and cluster analysis of docked poses. The RMSD calculation compares docked structures with a reference structure, also when atoms are randomly labelled, and their conformational and positional differences can be visualised. In addition, it is possible to visualise a ligand into the target binding pocket and investigate the key receptor–ligand interactions. Moreover, PacDOCK enables the clustering of docking results by identifying a restrained number of clusters from many docked poses. We believe that PacDOCK will contribute to facilitating the analysis of docking results to improve the efficiency of computer-aided drug design.

show abstract

Section: Resultsmentioning

confidence: 99%

PacDOCK: A Web Server for Positional Distance-Based and Interaction-Based Analysis of Docking Results

Carbone

Ghidini

Romano³

et al. 2022

Molecules

View full text Add to dashboard Cite

show abstract

“…DFS traverses a graph structure from an arbitrary starting node, exploring as far as possible along each branch then backtracking along all paths (Mercado et al 2021). In our hydrogen bond analysis tool, a random residue from the residue list is chosen as a starting node, then all the routes from this node that connect to QB are recorded.…”

Section: Openmm MD Simulationmentioning

confidence: 99%

Comparison of Proton Transfer Paths to the Q_Aand Q_BSites of theRb. sphaeroidesPhotosynthetic Reaction Centers

Wei

Zhang

Mao

et al. 2022

Preprint

View full text Add to dashboard Cite

The photosynthetic bacterial reaction centers from purple non-sulfur bacteria use light energy to drive the transfer of electrons from cytochrome c to ubiquinone. Ubiquinone bound in the QA site cycles between quinone, QA, and anionic semiquinone, QA•−, being reduced once and never binding protons. In the QB site, ubiquinone is reduced twice by QA&•−, binds two protons and is released into the membrane as the quinol, QH2. The network of hydrogen bonds formed in a molecular dynamics trajectory was drawn to investigate proton transfer pathways from the cytoplasm to each quinone binding site. QA is isolated with no path for protons to enter from the surface. In contrast, there is a complex and tangled network requiring residues and waters that can bring protons to QB. There are three entries from clusters of surface residues centered around HisH126, GluH224, and HisH68. The network is in good agreement with earlier studies, Mutation of key nodes in the network, such as SerL223, were previously shown to slow proton delivery. Mutational studies had also shown that double mutations of residues such as Asp M17 and AspL210 along multiple paths in the network presented here slow the reaction, while single mutations do not. Likewise, mutation of both HisH126 and HisH128, which are at the entry to two paths reduce the rate of proton uptake.

show abstract

“…Thus, by supplying a start symbol, new symbols can be sampled from the probability distribution corresponding to the next symbol (output by the RNN), which is then recursively fed back into the network resulting in de novo molecules. Despite a wave of newer approaches since (e.g., JT-VAE [ 13 ], DrugEx [ 29 ], GENTRL [ 30 ], GraphINVENT [ 20 , 31 ]), RNNs are still frequently used and investigated for de novo molecule generation (e.g., [ 32 – 34 ]). Furthermore, they still match the state-of-the-art on several de novo molecule generation benchmarks [ 22 , 23 , 35 , 36 ].…”

Section: Introductionmentioning

confidence: 99%

Augmented Hill-Climb increases reinforcement learning efficiency for language-based de novo molecule generation

et al. 2022

View full text Add to dashboard Cite

A plethora of AI-based techniques now exists to conduct de novo molecule generation that can devise molecules conditioned towards a particular endpoint in the context of drug design. One popular approach is using reinforcement learning to update a recurrent neural network or language-based de novo molecule generator. However, reinforcement learning can be inefficient, sometimes requiring up to 105 molecules to be sampled to optimize more complex objectives, which poses a limitation when using computationally expensive scoring functions like docking or computer-aided synthesis planning models. In this work, we propose a reinforcement learning strategy called Augmented Hill-Climb based on a simple, hypothesis-driven hybrid between REINVENT and Hill-Climb that improves sample-efficiency by addressing the limitations of both currently used strategies. We compare its ability to optimize several docking tasks with REINVENT and benchmark this strategy against other commonly used reinforcement learning strategies including REINFORCE, REINVENT (version 1 and 2), Hill-Climb and best agent reminder. We find that optimization ability is improved ~ 1.5-fold and sample-efficiency is improved ~ 45-fold compared to REINVENT while still delivering appealing chemistry as output. Diversity filters were used, and their parameters were tuned to overcome observed failure modes that take advantage of certain diversity filter configurations. We find that Augmented Hill-Climb outperforms the other reinforcement learning strategies used on six tasks, especially in the early stages of training or for more difficult objectives. Lastly, we show improved performance not only on recurrent neural networks but also on a reinforcement learning stabilized transformer architecture. Overall, we show that Augmented Hill-Climb improves sample-efficiency for language-based de novo molecule generation conditioning via reinforcement learning, compared to the current state-of-the-art. This makes more computationally expensive scoring functions, such as docking, more accessible on a relevant timescale.

show abstract

Exploring Graph Traversal Algorithms in Graph-Based Molecular Generation

Cited by 9 publications

References 23 publications

PacDOCK: A Web Server for Positional Distance-Based and Interaction-Based Analysis of Docking Results

PacDOCK: A Web Server for Positional Distance-Based and Interaction-Based Analysis of Docking Results

Comparison of Proton Transfer Paths to the Q_Aand Q_BSites of theRb. sphaeroidesPhotosynthetic Reaction Centers

Augmented Hill-Climb increases reinforcement learning efficiency for language-based de novo molecule generation

Contact Info

Product

Resources

About

Exploring Graph Traversal Algorithms in Graph-Based Molecular Generation

Cited by 9 publications

References 23 publications

PacDOCK: A Web Server for Positional Distance-Based and Interaction-Based Analysis of Docking Results

PacDOCK: A Web Server for Positional Distance-Based and Interaction-Based Analysis of Docking Results

Comparison of Proton Transfer Paths to the QAand QBSites of theRb. sphaeroidesPhotosynthetic Reaction Centers

Augmented Hill-Climb increases reinforcement learning efficiency for language-based de novo molecule generation

Contact Info

Product

Resources

About

Comparison of Proton Transfer Paths to the Q_Aand Q_BSites of theRb. sphaeroidesPhotosynthetic Reaction Centers