Nanopore sequencing, commercialized by Oxford Nanopore Technology (ONT), is a high-throughput genome sequencing platform. Unlike traditional sequencing-by-synthesis methods, nanopore sequencing uses measured current signals to sense the nucleotide sequence flowing through the pore. The signal-to-base conversion process introduces unique error patterns, making it challenging to design methods that rely on hand-crafted features. Deep learning uses multiple layers to progressively learn complex patterns in the input data, making it suitable for genome analysis. In this dissertation research, I present methods I developed based on deep neural networks to improve genome inference with nanopore sequencing. First, I introduce haplotype-aware variant calling pipeline PEPPER-Margin-DeepVariant that produces state-of-the-art results for nanopore long-reads. Next, I demonstrate a pipeline to perform de novo assembly of eleven human genomes in nine days. Then I show the application of the methods to validate and correct errors in the first complete human genome assembly. Finally, I demonstrate the utility of PEPPER-Margin-DeepVariant paired with highly multiplexed nanopore sequencing for rapidly identifying disease-causing variants.