English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Recently, pre-trained foundation models have enabled significant advancementsin multiple fields. In molecular machine learning, however, where datasets areoften hand-curated, and hence typically small, the lack of datasets withlabeled features, and codebases to manage those datasets, has hindered thedevelopment of foundation models. In this work, we present seven novel datasetscategorized by size into three distinct categories: ToyMix, LargeMix andUltraLarge. These datasets push the boundaries in both the scale and thediversity of supervised labels for molecular learning. They cover nearly 100million molecules and over 3000 sparsely defined tasks, totaling more than 13billion individual labels of both quantum and biological nature. In comparison,our datasets contain 300 times more data points than the widely used OGB-LSCPCQM4Mv2 dataset, and 13 times more than the quantum-only QM1B dataset. Inaddition, to support the development of foundational models based on ourproposed datasets, we present the Graphium graph machine learning library whichsimplifies the process of building and training molecular machine learningmodels for multi-task and multi-level molecular datasets. Finally, we present arange of baseline results as a starting point of multi-task and multi-leveltraining on these datasets. Empirically, we observe that performance onlow-resource biological datasets show improvement by also training on largeamounts of quantum data. This indicates that there may be potential inmulti-task and multi-level training of a foundation model and fine-tuning it toresource-constrained downstream tasks.

Towards Foundational Models for Molecular Learning on Large-Scale Multi-Task Datasets