Learning to embed lifetime social behavior from interaction dynamics - Data

Interaction matrices and metadata used in "Learning to embed lifetime social behavior from interaction dynamics"

The following files are included:

• interactions_bn16_sparse.npz and interactions_bn19_sparse.npz: These are the interaction affinity matrices for the BN16 and BN19 datasets as described in the publication. The data is stored as compressed sparse tensors with time on the first, and the individuals on the second and third dimensions. The data was stored using the pydata/sparse library 0.9.1

• alive_bn16.csv and alive_bn19.csv: These files contain the dates of emergence (also corresponding to the dates they were introduced into the colonies) and heuristically determined number days alive for all individuals in the interaction matrices. Death dates were determined using a bayesian changepoint model and the number of daily detections of each individual

• rhythmicity_bn16.csv and rhythmicity_bn19.csv: These files contain the circadian rhythmicity values used in the evaluation of the method. The circadian rhythmicity is the \(R^2\) value of a sine with a 24 hour period fitted to the individuals' movement velocities over a three day window

• indices_bn16.csv and indices_bn19.csv: These files contain the mapping between the original marker IDs used during the recording of the data (which has gaps, because not all markers were used) and the sequential indices used in the interaction matrices. These files can therefore be used to look up the original ID of an individual based on it's index in the interaction matrix and vice versa

• time_spent_on_substrates.csv: This data was used for the mapping from factors to the proportion of time spent on various cell substrates (Figure 5). The positions of the individuals were accumulated by minute, and the column "location_descriptor_count" contains the total number of minutes on the respective day that the individual was detected

See 10.1101/2020.05.06.076943 for more details about the bayesian changepoint model, circadian rhythmicity calculation, and location mapping.