MeSH terms), where those are missing. After assembling the dataset, we conduct experiments on link prediction in the co-annotation graph to analyze to what extent the research directions, expressed as the co-occurrence of MeSH terms, can be known ahead of time. In practice, link prediction in co-annotation graphs could be used for recommending promising research directions to researchers. Such applications are only possible because of the expanded view of our newly assembled dataset.In summary, our contributions are:• We provide a new dataset of COVID-19 publication data.• The dataset contains COVID-19 research papers along with first-order cited work. • We use ConceptMapper [2] to generate MeSH annotations, whenever those annotations are not present. • We conduct experiments on link prediction between concepts from the newly created dataset. • We describe the procedure for assembling the dataset and provide the code for keeping the data collection up-to-date. II. RELATED DATA COLLECTIONS We describe existing collections of COVID-19 research articles that are relevant to the dataset introduced in this work. a) CORD-19: COVID-19 Open Research Dataset 3 [1]. CORD-19 is a free and open dataset of research articles on COVID-19. It is maintained by the Semantic Scholar team at the Allen Institute for AI in collaboration with leading research groups. As of Aug 9, 2021, The dataset covers more than 280,000 scholarly articles. b) CrossRef: CrossRef has released a 65GB data file 4 to support COVID-19 research. The file contains 112M metadata records. These records are, however, not limited to COVID-19 2 https://www.nlm.nih.gov/mesh/meshhome.html 3 https://www.semanticscholar.org/cord19 4 https://www.crossref.org/blog/free-public-data-file-of-112-million-crossref-records/ Abstract-COVID-19 research datasets are crucial for analyzing research dynamics. Most collections of COVID-19 research items do not to include cited works and do not have annotations from a controlled vocabulary. Starting with ZB MED KE data on COVID-19, which comprises CORD-19, we assemble a new dataset that includes cited work and MeSH annotations for all records.Furthermore, we conduct experiments on the analysis of research dynamics, in which we investigate predicting links in a co-annotation graph created on the basis of the new dataset. Surprisingly, we find that simple heuristic methods are better at predicting future links than more sophisticated approaches such as graph neural networks.