liveBook · Manning

Reproducibility

As explained in Build AI Drug Discovery Pipelines

Reproducibility is the ability to obtain consistent results by repeating an analysis with the same documented inputs, software, models, and procedures.

Reproducibility means obtaining consistent results by repeating an analysis with the same documented models, datasets, preprocessing, parameters, and procedures. In end-to-end drug-discovery pipelines, it requires managing versions across every component. section 13.4.3

A reproducible analysis records:

  • The version of every component model, including changes in training data, architecture, or hyperparameters. section 13.4.3
  • The training datasets, preprocessing pipelines, and inference parameters used to produce results. section 13.4.3
  • End-to-end validation, because errors can propagate through cascaded structure-prediction and docking models. section 13.4.3

Changing a model version

An analysis run with ESM-2 version 1.0 may produce different results from the same analysis run with version 2.0. Reproduction therefore requires preserving the model version, training data, preprocessing, and inference settings rather than recording only the final output. section 13.4.3

How Build AI Drug Discovery Pipelines covers it

How deep it goes
Specifies tracking models, datasets, preprocessing, and inference parameters, while emphasizing validation rather than detailing an implementation checklist.
Examples
Compares analyses using ESM-2 versions 1.0 and 2.0 and considers cascaded structure-prediction and docking accuracy.
What you can do after
Operate reproducible analyses by recording component versions, datasets, preprocessing, inference parameters, and end-to-end validation results.
How it connects
Connects model versioning with training datasets, preprocessing pipelines, inference parameters, structure prediction, docking, and error propagation.

Also explained in

Compare how 11 books explain reproducibility →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Build AI Drug Discovery Pipelines. The section links point to the parts of the book it draws on.