Pushback on 'push‑button' bioinformatics

Bioinformatics is not a “push-button” job, Ming “Tommy” Tang said, arguing that biology still needs human judgment beyond standard software workflows. (x.com) Tang, AstraZeneca’s director of bioinformatics, made the case in a July 2026 thread on X. He said many teams still treat the field as if one default RNA sequencing pipeline can answer most questions. (x.com, divingintogeneticsandgenomics.com) RNA sequencing, or RNA-seq, reads which genes a cell is using by counting RNA molecules, then runs those reads through steps such as quality control, alignment, quantification, and differential-expression testing. Standard references describe that workflow as a multi-step process that starts with raw sequencing files and requires command-line tools and statistical analysis. (pmc.ncbi.nlm.nih.gov, bioinformatics.ccr.cancer.gov) Tang’s point was that this kind of pipeline is only the start. He said artificial intelligence in biology will need real-time biological data and tighter links between lab work, disease context, and computation than a one-click analysis can provide. (x.com) That argument lands as drug companies and research groups push more machine learning into genomics, pathology, and target discovery. In that setting, a model is only as useful as the data coming off instruments, the experimental design behind it, and the biological assumptions built into the analysis. (bioinformatics.ccr.cancer.gov, pmc.ncbi.nlm.nih.gov) Training materials for beginners make the same point in less blunt language. The National Cancer Institute’s 2025 RNA-seq course lists Unix, high-performance computing, quality control, alignment, counting, and interpretation as separate skills, not a single automated step. (bioinformatics.ccr.cancer.gov) Published RNA-seq guides also note limits on “basic computational analysis.” A widely used Current Protocols review says its standard workflow does not cover specialized cases such as single-cell RNA sequencing or other assay types that often need different methods and assumptions. (pmc.ncbi.nlm.nih.gov) Tang has spent years teaching those distinctions in public. His AstraZeneca profile says he works across computational biology, single-cell genomics, and cancer research, and his tutorials focus on tools such as Seurat, assay-specific methods, and reproducible workflows rather than one universal recipe. (divingintogeneticsandgenomics.com) The thread’s closing message was narrower than a rejection of automation. Tang said software can speed routine analysis, but he cast bioinformatics as a mix of biology, statistics, coding, and fresh data that still resists a single button. (x.com)