v0.1.4 | 7 format adapters | 83 conformance tests | Evidence bundles | Numba JIT | CLI | OS v4.0
BioSDK is a unified library for biological neural data. One open() for any format — MEA, EEG, ecephys, Sleep, RNG — with integrity-verified evidence bundles. Think of it as pandas.read_* for neural data.
- One
open(), 7 formats: HDF5 (MCS, Giroldini), NWB (DANDI/Allen), EDF (PhysioNet, OpenNeuro), XLSX (Tressoldi), CSV (GCP2), FinalSpark API. - Same features across vendors: 6 standard features per channel (RMS, MAV, zero-crossings, variance, peak, skewness) — compare MEA from MCS to MEA from Giroldini in one pipeline.
- Evidence bundles: Every result is a SHA256-chained, HMAC-signed bundle — manifest + data + results + signatures in one folder. Reproducible, verifiable, tamper-evident. This is unique to BioSDK — no other neural data library provides this.
- Open Core: MIT (community) + Commercial (enterprise safety/closed-loop).
| Tier | License | Includes | Price |
|---|---|---|---|
| Community | MIT | Format adapters, features, readout, evidence bundles, dashboard | Free |
| Enterprise | Commercial | Closed-loop safety controller, managed cloud, SLA, priority support | Contact |
Why Open Core: The library must be maximally adopted — MIT ensures zero friction. Revenue comes from enterprise features that labs and companies need for production.
# From real PyPI:
pip install biosdk
# From TestPyPI (pre-release):
pip install -i https://test.pypi.org/simple/ biosdkOptional extras:
pip install "biosdk[dashboard]" # Web dashboard
pip install "biosdk[mne]" # EEG/EDF support via MNE
pip install "biosdk[all]" # Everythingimport biosdk
# Open ANY neural data — auto-detects format (HDF5, NWB, EDF, XLSX, CSV)
ds = biosdk.open("recording.h5")
# 6 standard features per channel: RMS, MAV, ZC, VAR, PEAK, SKEW
X = biosdk.features(ds, window_s=1.0)
# Classification readout (Random Forest, Logistic Regression, SVM)
result = biosdk.readout(X, y, classifier="rf")
# Integrity-verified evidence bundle (SHA256-chained, HMAC-signed)
bundle = biosdk.evidence_bundle(result, "my_results")| # | Adapter | Format | Vendor | Modality | Ch | Hz | Tests |
|---|---|---|---|---|---|---|---|
| 1 | mcs_mea2100 | HDF5 | MCS | MEA | 17 | 500 | 8/8 |
| 2 | dandi_nwb | NWB | DANDI/Allen | ecephys | 5-96 | 100-30k | 8/8 |
| 3 | physionet_edf | EDF | PhysioNet | Sleep/EEG | 7-21 | 100-200 | 8/8 |
| 4 | gcp2_csv | CSV | GCP2 | RNG | 1 | 1/60 | 9/9 |
| 5 | tressoldi_h3 | XLSX | Tressoldi | EEG | 14 | 128 | 9/9 |
| 6 | finalspark_neuroplatform | API | FinalSpark | MEA wetware | 8 | 30k | 41/41* |
| R | giroldini_mea | HDF5 | Giroldini | MEA | 59 | 20k | ref |
* FinalSpark: adapter skeleton built. Conformance tested on mocks (41/41). Awaiting API token for live validation.
Conformance total: 42 certified + 41 mock = 83/83 PASS
All numbers are current-snapshot values. Chance rates shown for context.
| Dataset | Task | Accuracy | Chance | Improvement |
|---|---|---|---|---|
| OpenNeuro ds007558 | Eyes open/closed (2-class) | 83.7% | 50% | 1.7x |
| Sleep PSG (PhysioNet) | Sleep staging (5-class) | 66.7% | 20% | 3.3x |
| Tressoldi H3 BBI | Stimulus vs rest (2-class) | 64.2% per-pair | 50% | 1.3x |
| Giroldini MEA | 4-class stimulus | 52.4% | 25% | 2.1x |
| Cross-vendor MEA | Giroldini vs MCS | 100% separable | 50% | 2.0x |
Honest baseline: sklearn SVM achieves 50.7% on the same Giroldini MEA task. BioSDK's pipeline adds +1.7 percentage points. The value is in the unified API + evidence bundles, not in algorithmic edge over standard classifiers.
NSI-1.0 features preserve modality identity — same modality clusters together, different modalities separate:
| Modality cluster | Correlation | Interpretation |
|---|---|---|
| MEA (Giroldini-MCS-DANDI) | 0.90-0.96 | Same modality, different vendors |
| EEG (OpenNeuro-Sleep) | 1.00 | Same modality, same format (EDF) |
| Cross-modality (MEA vs EEG vs RNG) | ~0.00 | Different physical phenomena — correct behavior |
- ❌ NOT a new file format (NWB, EDF, HDF5 already exist — BioSDK is a unified loader, not a replacement)
- ❌ NOT an operating system ("BiC OS" was an early working name — the project is a Python library)
- ❌ NOT a biological computer, GPU replacement, or energy platform
- ❌ NOT production-hardened for live stimulation (closed-loop safety gates are designed and simulator-tested; hardware validation pending FinalSpark token)
BioSDK includes BioReservoirV40 — a biologically-plausible spiking reservoir (Izhikevich neurons, small-world connectivity, STDP plasticity) with Numba JIT acceleration (1.3x speedup). Processes real neural spike data through recurrent dynamics and beats sklearn on temporal tasks.
| Task | BioReservoir | sklearn | Delta |
|---|---|---|---|
| 4-MEA temporal classification | 63.79% | 29.18% | +34.6pp |
| 8-MEA temporal classification | 48.33% | — | (3.87x chance) |
| 42-MEA temporal classification | 🔄 sweep running | — | — |
Key finding: STDP plasticity A+ has inverted U-curve — A+=0.01 is optimal, A+>0.02 destroys generalization. FS-heavy neuron distribution + input_scale=10.0 gives best results on 8-MEA.
biosdk open recording.h5 # Show metadata
biosdk features recording.h5 -w 0.5 # Extract features
biosdk readout recording.h5 labels.txt # Classify
biosdk evidence results.json # Evidence bundle
biosdk os start|stop|status|jobs # OS control
biosdk benchmark # Reservoir health
biosdk list # List adapters
biosdk version # System infoProduction-grade operating system layer with:
- RBAC: 4 roles (admin/operator/researcher/viewer), 9 permissions, API key + HMAC session auth
- Lab Approval: 4-stage workflow (submit→review→execute→complete) with 7 safety gates
- Evidence Ledger: SHA256-chained append-only tamper-evident audit trail
- Daemon + heartbeat, JobScheduler (4 workers), DeviceManager, Telemetry
Live web dashboard with 5 pages: OS overview, reservoir visualization (membrane potentials heatmap + spike counts), job queue, device manager, safety gates.
python -m biogpu.dashboard.server_v40
# → http://127.0.0.1:8420| # | Item | Status |
|---|---|---|
| 1 | BioReservoir sweep on 42-MEA | 🔄 Running (Phase B) |
| 2 | GPU/Energy comparison framework | Planned |
| 3 | OS permissions + lab approval | ✅ Done (v4.0) |
| 4 | BioSDK CLI + Docker | ✅ Done |
| 5 | CI/CD GitHub Actions | Planned |
| 6 | FinalSpark live validation | Awaiting token |
| 7 | Closed-loop on BioReservoirV40 | ✅ Done (v32) |
| 8 | Multi-timescale reservoir (if 42-MEA < 15%) | Planned |
| 9 | PyTorch CUDA batch reservoir | Planned |
| Attribute | Value |
|---|---|
| Name | BioSDK (BioCompute Software Development Kit) |
| Positioning | Unified neural data library + evidence bundles |
| Package | biosdk v0.1.4 |
| License | Open Core: MIT (community) + Commercial (enterprise) |
| Author | Vladislav Dobrovolskii (vladimoryachok@gmail.com) |
| PyPI | https://pypi.org/project/biosdk/ |
| GitHub | https://github.com/Vladrus39/BioSDK |
| Document | Path |
|---|---|
| Session Handoff (v4.3→v4.4) | HANDOFF_FOR_DEEPSEEK_2026_05_11.md |
| Session Start Prompt | SESSION_START_PROMPT.md |
| Master Project Plan | docs/MASTER_PROJECT_PLAN_V4.md |
| Master Status | MASTER_STATUS_V85.md |
| Machine Status | PROJECT_STATUS_V85.json |
| NSI-1.0 Spec | docs/standards/NSI_1_0_SPECIFICATION.md |
| Evidence Bundle Format | docs/BIOSDK_EVIDENCE_PACK_GUIDE_V512.md |
| Numba JIT Module | biogpu/substrates/bio_reservoir_numba.py |
| CLI | biosdk/cli.py |
| OS Permissions | biogpu/production/permissions_v40.py |
| Lab Approval | biogpu/production/lab_approval_v40.py |
| Evidence Ledger | biogpu/production/evidence_ledger_v40.py |
- Community Edition (this repository): MIT License — adapters, features, readout, evidence bundles, dashboard
- Enterprise Edition: Commercial License — closed-loop safety controller, managed cloud, SLA
BioSDK v0.1.4. Unified neural data library. Open Core. Evidence bundles. One open() for 7 formats.