STdeconvolve as an unsupervised, reference-free approach to infer latent cell-type proportions and transcriptional profiles within multi-cellular spatially-resolved pixels from spatial transcriptomics (ST) datasets. STdeconvolve builds on latent Dirichlet allocation (LDA), a generative statistical model commonly used in natural language processing for discovering latent topics in collections of documents. In the context of natural language processing, given a count matrix of words in documents, LDA infers the distribution of words for each topic and the distribution of topics in each document. In the context of ST data, given a count matrix of gene expression in multi-cellular ST pixels, STdeconvolve applies LDA to infer the putative transcriptional profile for each cell-type and the proportional representation of each cell-type in each multi-cellular ST pixel.