Visualization of chemical compounds in graphs and plots is a challenging task, due to the high complexity of data and high number of dimensions of its machine representation. Such visualizations should be easily interpretable. Unsupervised learning techniques such as PCA, t-SNE or MDS can be used to resolve dimensionality issues but may lead to overlapping or widely scattered points, making interpretation difficult. To solve the datapoints distribution problems, algorithms based on artificial forces can be used to spread out overlaps (by pseudo-spring force) and bring distant points closer (by pseudo-gravity). 

In this study we explored ForceAtlas2 algorithm [1] to visualize materials data as graphs. We used attribute descriptor MEGnet [2] to transform chemical compound formulas into high-dimensional vectors. We investigated how ForceAtlas2 parameters and the rules for creating edges between graph nodes influence the graph. Our results indicate that force-based graphs can help improve placement of datapoints while maintaining chemical knowledge: on small scale by keeping similar compounds together, and large scale by forming clusters of different compound groups. Initial findings demonstrated on several datasets [3-5], highlight the potential for broader application in visualizing material properties.