This program is intended to take sequencing data from an aptamer binding experiment (see Arter et al, Anal. Chem., 2018, 90 (17), pp 10302–10310), process it to identify sequences that successfully bound, and train an Ising model on these data. It consists of three parts:
Part I: Pre-processing Trim adaptors, remove low quality sequences One-hot encode Standardise Optionally: Augment FASTA with shifted sequences
Part II: Model training This trains a model on the encoded data options: Independent sites Mean-field ising Convolutional neural net Fully-connected neural net
Part III: Validation This visualises the results of the model training Generate samples from model (and compare to originals) Plot inferred coefficients on histogram Evaluate the probabilities of a set of sequences coming from the model
Prediction This returns a suggested probability distribution to be used to generate new inputs (the marginal distribution for sequences that bind)