thebench.bio #1 Ecosystem for the Bio-Tech Industry

Software Daily Summary

Software Daily Summary: What Happened on November 18, 2026

Antibiotic-resistance AI needs better data and tougher tests

NF-core set migration rules for typed Nextflow modules as an Infinitus executive described where AI agents could fit in pharma and health care.

  • Updated
  • Read from 6 articles, 6 publications
  • N P P

Today in brief

A 2024 comparison found that ResFinder performed better against machine-learning models when test bacteria were less related to the training strains.1 Jonathan L. Jacobs argues for authenticated strains, paired genome and susceptibility data, and shared benchmarks for resistance-prediction tools.1

NF-core published guidance for moving Nextflow modules to typed inputs and outputs, with minimum software versions tied to fixes in process caching.2 Infinitus CEO Ankit Jain said AI agents could take on routine health-care work, while people handle unusual cases and deeper investigations.3

Antibiotic-resistance AI needs reliable isolates and shared benchmarks

A 2024 comparison of four machine-learning models and ResFinder across 78 species-antibiotic datasets found that test-set design affected which approach performed better.1 When test strains were less related to training strains, ResFinder outperformed the machine-learning models on 44% to 50% of datasets, versus 25% with random splits.1

Jonathan L. Jacobs, ATCC’s bioinformatics director, argues that resistance models need high-quality lab results alongside genome sequences.1 He recommends authenticated bacterial strains with genome sequences and minimum inhibitory concentration (MIC) results from the same isolate.1

MIC measurements retain more detail than susceptible-or-resistant labels, which can become outdated when clinical breakpoints change, Jacobs writes.1 A 2020 report identified more than two million contaminated GenBank records, which could expose models to DNA assigned to the wrong organism.1

Jacobs calls for shared test sets with fixed data splits, updated to include recent clinical isolates and historical strains.1 ATCC researchers observed resistance genes accumulating over time when they compared historical and recent Acinetobacter baumannii isolates.1

NF-core sets migration rules for typed Nextflow modules

NF-core published guidance for updating modules to use Nextflow’s static typing and record types, described as mostly stable since version 26.04.0.2 The guidance calls for each module to accept and return one structured record, replacing the earlier allowance for multiple inputs and outputs.2

The guidance recommends Nextflow 26.10.4 or later for typed functions that return records, after that release fixed inconsistent record hashes that blocked process resumption.2 For modules that stage scripts, it recommends version 26.10.6 or later, which fixed staged files being left out of process hashes.2

The updated structure puts staged inputs and scripts in a process stage block and moves version information into a record in the topic block.2 It also removes ext.when from statically typed modules, leaving workflow developers to manage conditional logic.2

Infinitus CEO sees agents taking on routine pharma work

Infinitus CEO Ankit Jain said agents could handle routine health-care work, with people stepping in for unusual cases, deeper investigations, or regulatory needs.3

Jain said agents should help field teams and physicians get information and communicate, rather than resemble automated phone systems that frustrate users.3

Also worth knowing

  • A November 17 survey found gaps in pharmaceutical companies’ internal AI oversight and uncertainty about who is accountable for AI systems.4

  • A November 17 perspective argued that biopharma companies should fund tools to measure biological features current assays miss, because more data from existing assays may leave their blind spots intact.5

  • P

    A November 18 commentary argued that AI can help drug researchers ask deeper questions, not just supply answers.6

Also published