Start with a useful public catalogue, then make small, well-documented datasets and learning recipes reproducible. Maintain source provenance and licences separately from the code that downloads or transforms them.
Students, independent developers, researchers and teams building their first biomedical model.
First product
A resource catalogue that explains what to learn or train, how to obtain the material, and what can be reused or forked.
Next decision
Every listed resource has a source link, use case and explicit acquisition and licence notes.
Product boundary
A publicly accessible resource is not automatically unrestricted training material. Catalogue entries link upstream; no complete database mirror or OpenPharma-trained model is claimed.
PUBLIC LEARNING & TRAINING RESOURCES
Start with what is already open.
Authoritative datasets, software and model weights for learning and research. This catalogue links upstream materials; availability here does not mean OpenPharma has mirrored or validated them.
Source checked 25 September 2026. No local data mirror or completed OpenPharma model is claimed.
Fork code. Version data. Keep the provenance.
A code fork preserves software history. For large databases, publish a selection manifest and reproducible download recipe instead of putting a whole database into Git. Before sharing a snapshot, review the specific data terms and retain attribution, identifiers, release dates and checksums.
For models, track code, training data and checkpoint permissions separately. A code licence does not automatically license every dataset or weight file used by it.
Three starting learning paths
Molecular features: select a small PubChem set, use RDKit to calculate descriptors, and record every input and transformation.
A prediction baseline: choose one ChEMBL assay task, use a simple DeepChem baseline, and compare appropriate held-out splits.
Target evidence: connect UniProt identifiers with Open Targets evidence and inspect the provenance of each association.
These are proposed exercises, not tested OpenPharma tutorials. Training and hosted execution are planned for later phases.
PRODUCT 0 → 1.0
The Phase 0–5 strategy.
Each phase has a concrete deliverable and an acceptance gate. These are proposed development plans, not completed scientific or product validations.
PHASE 0 / 0% MILESTONECurrent planning
Publish the starting catalogue
Make existing materials easier to discover and evaluate.
Work to do
List authoritative datasets, toolkits and model checkpoints.
Explain useful learning tasks, access methods and reuse terms.
Distinguish an upstream resource from a locally mirrored or tested asset.
Acceptance gate
Every listed resource has a source link, use case and explicit acquisition and licence notes.
Publish: Public resource catalogue; product progress remains 0% until the scope is accepted.
PHASE 1 / 20% MILESTONEPlanned
Create reproducible manifests
Make selections repeatable without copying entire databases into Git.
Work to do
Record upstream release or commit, retrieval date and identifiers.
Define checksum, transformation and attribution fields.
Choose a small permitted subset for the first lesson.
Acceptance gate
A second person can identify the exact intended inputs and their permitted reuse.
Publish: Manifest schema and a reviewed starter selection.
PHASE 2 / 40% MILESTONEPlanned
Publish the first learning recipe
Let a learner complete one realistic task from a clean setup.
Work to do
Write an acquisition script with limits, retries and clear errors.
Provide an environment lock and a notebook with expected outputs.
Use a public reference task such as molecular featurisation.
Acceptance gate
A clean-machine rerun produces matching data checksums and expected outputs.
Publish: Versioned lesson, script and reproducibility record.
PHASE 3 / 60% MILESTONEPlanned
Evaluate datasets and models
Show quality and limitations before people invest in training.
Work to do
Add dataset cards for missingness, label quality and leakage risks.
Add model cards distinguishing checkpoints from training code.
Run benchmarks with fixed splits and publish compute requirements.
Acceptance gate
Independent reruns meet declared tolerances and all required cards are complete.
Publish: Reviewed dataset/model cards and benchmark bundles.
PHASE 4 / 80% MILESTONEPlanned
Pilot community replication
Learn where setup and documentation fail for other people.
Work to do
Invite a small learner and researcher cohort to follow the recipes.
Record successful and failed replications with environment details.
Triage issues and make fixes visible in the changelog.
Acceptance gate
The predefined replication target is met across documented environments.
Publish: Community replication report and improved lessons.
PHASE 5 / 100% MILESTONEPlanned
Release the shared workspace
Maintain a dependable catalogue and recipe collection.
Work to do
Publish stable manifests, version history and update ownership.
Support resource search, exports and documented contribution review.
Offer managed execution only after cost, security and data-rights review.
Acceptance gate
Supported recipes pass repeatability checks and release maintenance is funded and assigned.
Publish: Product 1.0 resource and replication workspace.
Progress stays at 0% until the starting scope is accepted and subsequent milestone evidence is published. Phase 1–5 targets are 20%, 40%, 60%, 80% and 100%. These percentages track development, not treatment effectiveness.
WHAT THIS DEPENDS ON
Build with the right foundations.
Upstream maintainers: authoritative documentation, releases and reuse terms.
A reproducibility reviewer: run examples independently on a clean machine.
Discovery and repurposing modules: define the first practical learning tasks.
Partner roles above are requirements. No partner participation or endorsement is claimed.
FUTURE INTERACTIVE PRODUCT
Dataset & model workspace
The public resource catalogue is available now. Managed downloads, hosted training and experiment execution remain planned.
Search resources by task and material type.
Review reuse terms and acquisition instructions.
Select a pinned recipe and reproduce it locally.
Compare checksums, environment and benchmark results.
Created a public code fork of rdkit/rdkit with the upstream BSD 3-Clause licence retained. This establishes a software starting point; no dataset mirror, benchmark result or completed product milestone is claimed.
Created a public code fork of deepchem/deepchem with its upstream MIT licence retained. No OpenPharma model training, dataset mirroring or benchmark validation has been completed.
Added ten upstream datasets, software projects and a model checkpoint with access instructions and reuse notes. These are learning and research resources, not a locally validated training collection or a complete database mirror.
Published the module strategy, concrete phase deliverables, acceptance criteria and future portal workflow. This is a planning update; product completion remains 0%.
Entries are published with site updates. Planned work is labelled separately from completed product milestones.