Identifying overrepresented concepts in gene lists from literature: a statistical approach based on Poisson mixture model