Authors: Maciej Sypetkowski, Frederik Wenkel, Farimah Poursafaei, Nia Dickson, Karush Suri, Philip Fradkin, Dominique Beaini

Published on: April 17, 2024

Impact Score: 7.6

Arxiv code: Arxiv:2404.11568

Summary

• What is new: The first observation of GNNs showing significant improvements from scaling up, using the largest collection of 2D molecular graphs to date.
• Why this is important: GNNs have struggled to demonstrate the benefits of scaling due to sparse operations inefficiency and unclear architectural effectiveness.
• What the research proposes: Evaluating different GNN architectures, including message-passing networks, graph Transformers, and hybrids, on a massive dataset to study scaling effects.
• Results: Achieved a 30.25% performance improvement with 1 billion parameters, and a 28.98% improvement with an eightfold increase in dataset size, showing strong finetuning scaling behavior on 38 tasks.

Technical Details

Technological frameworks used: nan

Models used: GNN architectures, including message-passing networks, graph Transformers, and hybrid models

Data used: The largest public collection of 2D molecular graphs

Potential Impact

Pharmaceutical companies, particularly those involved in drug discovery, could experience significant disruption or benefit from these insights.

Want to implement this idea in a business?

We have generated a startup concept here: MolecuNet.