Improving de novo sequence assembly using machine learning and comparative genomics for overlap correction