High-quality disambiguation is required to correctly link researchers to their grants and outputs including articles, patents, and clinical trials. The NIH Office of Portfolio Analysis developed a disambiguation solution that used article level metadata to assign 24.5M unique papers from the PubMed database to 16.0M unique author names, then used a novel neural network model trained on ORCID identifiers to determine whether author-publication pairs refer to variant representations of the same person. For example, our model can determine whether hypothetical records listing Jane Smith and Jane M. Smith were the same person, or two different people, based on variables that include institutional affiliation, co-authorship, and article-affiliated Medical Subject Heading (MeSH) terms. For more information see the publication describing this method: Yu et al 2021 (https://www.biorxiv.org/content/10.1101/2021.02.02.429450v1.full.pdf)