A literature base for a DIY bio-fungal brain–computer interface: neurons interfaced with a mycelium network, digitized onto a software layer where real work can be done — with initiation and evolution protocols, honest costs, power budgets, and the cheapest credible way in. It is no longer a solo build: it now runs as a two-person R&D loop — an archwizard on the systems and software side, an accredited chemical engineer on the chemistry and wet-lab side.
Built from 54 published systems across six domains: brain–computer interfaces, organoid intelligence, fungal mycelium networks, in-vitro neuronal networks, membranes & substrates, and hybrid systems. Every price is tagged [published] [reported] [estimated] or [unknown], exactly as the underlying research found it. Prices are index-based (vendor pages crawled within 1–14 days of the research date) — re-check before buying.
This stopped being a solo build. The project now runs as a two-person R&D team: a technomancer who builds the systems and the digitization layer, and an accredited chemical engineer who brings the chemistry, the biology, and wet-lab discipline. Every protocol below carries an owner, and the two halves meet where it matters most — the hybrid junction experiments, which neither seat can run alone.
A scholar-technomancer: inventing magic with technology. Maxx holds the vision and the architecture — how living tissue becomes data, and what the combined system is for.
An accredited, educated chemical engineer, with research experience on the amygdala and a working interest across neurology, biology, and chemistry. He is the credentialed half of the bench — the reason the neuronal side of this project is now realistic rather than aspirational.
The junction. A co-culture experiment is simultaneously a chemistry problem (shared medium, permeable barrier, interface materials — Jake) and a measurement problem (electrodes, acquisition, coupling statistics — Maxx). Phases 5–7 are deliberately co-owned, and no phase gate closes on one person's say-so: the engineer checks the wizard's data pipeline, the wizard checks the engineer's culture claims, and the record holds both.
One repeating cycle, run weekly, with a hard gate between phases. The loop is the real instrument: it is what turns two people's spare hours into compounding evidence instead of parallel hobbies — and it is the structural answer to this field's replication problem (§13), because every result passes a second, differently-trained pair of eyes before it is believed.
The experiment is written down before anything is built: a falsifiable hypothesis, the controls, and what number would settle it. Either seat can propose. Jake stress-tests the chemistry and biology; Maxx stress-tests whether the measurement can actually see the claimed effect.
The phase owner runs the work. The other seat reviews the setup before the run starts — sterile technique, electrode contact, code path — because a second pair of eyes is the cheapest control in the whole project.
Data and bench notes land in the record the same day: NWB/CSV for signals, a written log for wet work, everything indexed. If it isn't recorded, it didn't happen — the data layer is what makes the work fundable and publishable later.
A weekly joint session: data against hypothesis, anomalies first. Each result gets one of three verdicts — replicate it, extend it, or kill it — and the next experiment is chosen there, not mid-week on impulse.
A phase closes only when its exit criterion (below) is met and both seats sign off. The next phase's budget unlocks at the gate, not before — this is what keeps a two-person budget from diffusing across seven half-finished phases.
| Phase | Lead | The other seat |
|---|---|---|
| 0 · Skills & sterile technique | Split | Jake owns the bench (sterile technique, agar work); Maxx owns the software stack setup — both finish before Phase 1 starts |
| 1 · First digitized fungal signals | Both | Jake: growth, substrates, chamber chemistry · Maxx: recording rig, digitization, spike statistics |
| 2 · Fungal computing & sensing | Maxx | Jake runs the PEDOT:PSS-infusion chemistry (§54) and sensor-materials work |
| 3 · Human EEG (learn BCI on yourself) | Maxx | Jake reviews the pipeline and the safety rules (battery power only) before first use |
| 4 · Access & first neurons | Jake | Maxx supports the lab / university outreach and sets up the neuronal data plan in advance |
| 5 · First digitized neuronal signals | Both | Jake: cultures, plating, viability · Maxx: MEA acquisition, spike sorting, burst statistics |
| 6 · Closed-loop learning | Maxx | Jake keeps cultures healthy through training and owns the control conditions |
| 7 · The hybrid | Both, jointly | The junction experiments — co-culture chemistry and coupling measurement designed together, run together, judged together |
Neurons spike in milliseconds. Fungal spikes last minutes to hours and propagate at ~40 cm/h (~0.11 mm/s, §28) — roughly 40-million-fold slower than neuronal conduction. Any hybrid must treat mycelium as a slow modulatory / context layer, never a peer spike network. Close the loop at the burst-envelope level (seconds) on the neuronal side and the spike-train level (minutes–hours) on the fungal side.
Log-scale illustration from published values in §21 and §28. Signal amplitudes, by contrast, are compatible: fungal extracellular spikes ~0.5 mV (§24, §28), intracellular 5–50 mV (§21) — big enough for cheap 24-bit electronics. It's the speed, not the size, that doesn't match.
Mycelium's near-term value is as a living multi-modal sensor — moisture, pressure, chemistry, temperature (§31) — and a slow state layer feeding the fast system.
The neuronal culture does the millisecond work. Best architecture: reservoir computing — biology as a rich nonlinear reservoir, cheap silicon readout trained on top (§15).
Adamatzky's hypothesis (§50) is explicitly falsifiable via co-culture + MEA. A positive coupling measurement — or a bounded null — is the publishable result.
Ranges, not single numbers. Assumptions: no institutional overhead, community-lab membership where stated, builder's time unpaid (the dominant real cost), consumables at list price. Excluded: LN₂ storage setup [unknown], microscope (budget $500–2,000 used [estimated, unsourced — verify]), cell shipping/cold-chain, and failed cultures (budget 2–3× cell cost for learning-curve losses).
Low / mid points from the research: fungal $200 / $500 · neuronal $6,000 / $12,000 · platform $15,000 / $35,000. Rent-the-biology route (FinalSpark): $500/mo (2024, academic) to $1,000/mo + $1,000 setup; 2026 pricing is contact-only. CL1 cloud ≈ $300/week [reported]. Neither commercial route has any mycelium capability.
| Component | Draw | |
|---|---|---|
| CO₂ incubator (150–165 L) | ~100–300 W avg | [estimated] |
| Laptop / PC (acquisition, sorting) | 60–150 W | [estimated] |
| MEA2100-class amplifier | ~20–50 W | [estimated] |
| Intan RHD USB system | ~10–20 W | [estimated] |
| Mycelium chamber (mister/fan/LED) | ~5–15 W | [estimated] |
| Raspberry Pi 5 | ~7 W | [reported] |
| CL1 unit (reference) | ~30 W | [reported] |
Starter total (fungal + neuronal tiers, incubator via community lab): ~90 W continuous-equivalent ≈ 2.2 kWh/day. With your own incubator: ~200–400 W ≈ 5–10 kWh/day ≈ $22–45/month at an assumed $0.15/kWh (verify local rate). The incubator dominates by an order of magnitude — the strongest argument for the community-lab route. Incubator max ratings are 376–900 W [published]; averages are thermostat-cycled estimates.
Build in order of increasing cost and decreasing reversibility: fungi first (cheap, forgiving, BSL-1), human EEG second (learn the software stack on yourself), neurons last (expensive, fragile, regulated). Every phase ends with digitized signals and a publishable artifact. Each phase now carries its owner — Maxx systems & software, Jake chemistry & wet lab, Both where the work is genuinely shared — and closes only at the joint gate described in the R&D loop above.
Still-air box or DIY flow hood, pressure cooker as autoclave, PDA plates, agar transfers. Grow oyster mycelium on agar and grain. Read Adamatzky's methods (§24) and the DishBrain paper (§13) end to end. Set up Python, BrainFlow, SpikeInterface, LSL on a laptop.
Exit: clean agar plates, 3 contamination-free transfers in a row.
Oyster grain spawn ($20) on hemp pads / straw in a humidity-controlled tub. Recording rig: Raspberry Pi 5 4GB ($110 [published]) + ADS1256 24-bit ADC (~$13) + stainless needle electrodes (~$30 [estimated]). Record at 1 Hz, 24-bit, differential pairs — Adamatzky's stack (§24) at ~1/30th the cost of his $809 Pico ADC-24. Log CSV → HDF5/NWB. Replicate the thermal-stimulus cluster experiment (§22).
Exit: 7+ days continuous recording with spontaneous spikes (0.3–1 mV, minutes long) and a stimulus response; spike-train statistics computed.
8–16 electrode arrays on colonized blocks; impedance spectroscopy for colonization monitoring (§32); stimulus–response mapping (moisture, pressure, temperature, light) into a sensor calibration (§31); FHN-model gate mining (§25); the PEDOT:PSS-infusion experiment (§54).
Exit: a characterized mycelium sensor with a calibration curve, and one mined logic gate — or a documented failure with numbers.
PiEEG-8 ($350 [published]) + Pi 5 + electrodes/cap (~$100 [estimated]). Build a P300 speller (§10) and SSVEP demo (§11) with BrainFlow; record your own motor imagery (§12). The pipeline, LSL streaming, classification and closed-loop software built here is reused verbatim for the neuronal rig. Battery power only — never mains on a person.
Exit: working P300 speller at >80% accuracy and an SSVEP demo.
Join a community lab with a tissue-culture room (~$100/mo [reported]) or secure a university collaborator — this solves BSL-2 space, incubator access, and the institutional ordering barrier in one move. Order primary rat cortical neurons (Gibco $908 / Lonza $822 [published]) + B-27 Plus system ($303) + poly-D-lysine ($157). Plate, feed (half-media changes every 2–3 days), learn microscopy. Do not buy an incubator yet — use the lab's.
Exit: a healthy 14-DIV cortical culture with visible networks.
Build the Piphys-class rig (§36): Intan RHD USB interface ($2,755) + RHD2132 headstage ($895) or the $350–450 DIY headstage (§42); MCS 60-electrode TiN MEA chips (€306 each). Record spontaneous activity; spike-sort with SpikeInterface + Kilosort4.
Exit: sorted single units with stable waveforms over a 30-minute session; raster plots and burst statistics.
Shahaf & Marom contingent training (§33) first, then the DishBrain Pong protocol (§13) with an RHS2116 stim-capable headstage. Metrics: rally length, ace rate, long-rally fraction. Run controls (medium-only, open-loop) from the start — given the field's replication controversy, your controls are your credibility.
Exit: a statistically significant learning effect with controls — or a documented null with power analysis. Either is publishable.
Execute the Adamatzky hypothesis test (§50): neuron–mycelium co-culture on MEA behind a permeable barrier (shared medium, no cell mixing) → direct-contact co-culture → PEDOT:PSS-infused mycelium bridge (§54). Test both directions: does neuronal stimulation modulate fungal spiking? Does fungal activity modulate neuronal firing? Frame it as a coupled reservoir (§15) and train silicon readouts on combined features.
Exit: the first quantitative measurement of neuron–fungus electrophysiological coupling — or its absence, with bounds.
Ownership splits along the two clocks: Jake owns the biology — culture schedules, media chemistry, viability calls — while Maxx owns the measurement of maturation (impedance tracking, coupling statistics, readout training). The training ladder itself is joint work: neither seat declares “functional coupling” alone.
Neurons: plating → neurite outgrowth & first synapses (days 3–7) → first synchronized bursts (days 10–14) → mature bursting (days 14–21) → peak experimental window weeks 3–8 → decline over months 2–6 (peptide hydrogels extend viability to ≥5 months, §45; FinalSpark organoids ~100 days, §16; CL1 ~6 months, §17). Plan experiments in the day-14–60 window; keep 2–3 cultures staggered a week apart.
Mycelium: spawn run 10–12 days at 22–28 °C → full colonization in 2–4 weeks → electrically active for months with humidity maintained; dehydrated blocks store 6–12 months and revive (§31/§32). Mycelium is the long-lived partner; neurons are the short-lived one — design the fungal layer to persist across neuronal generations.
What was built, what it showed, and what to steal — condensed from the research report, with the primary source on every card. Filter by domain or search by name, technique, or number.
| # | Technique | From | Why first |
|---|---|---|---|
| 1 | Needle-electrode + 24-bit logger fungal recording at 1 Hz (reproducible ~$25 with ADS1256) | §24 | Cheapest digitized living signals in existence; week-1 build |
| 2 | DishBrain closed-loop recipe: predictable stim = reward, unpredictable = punishment; 8-electrode sensory encoding; rally/ace metrics | §13 | Fully published training protocol; needs no HD-MEA |
| 3 | Reservoir computing: biology as nonlinear reservoir + trained silicon readout | §15 | Sidesteps “does the dish learn?”; works with any rich dynamics, incl. mycelium |
| 4 | Intan RHS2116 + Open Ephys + RHX stack; DIY headstage with FPC connector | §42, §36 | Only affordable open path to bidirectional spike recording |
| 5 | Piphys: Pi + Intan, $819 / 32 ch, published noise figures | §36 | Reference DIY rig design |
| 6 | Shahaf & Marom contingent-stimulus-removal training | §33 | Simplest closed-loop learning protocol; one threshold detector |
| 7 | PiEEG-8 as universal biosignal front-end (ADS1299, 16 kSPS) | §3 | Cheapest credible EEG/EMG/ECG + slow-signal board |
| 8 | PEDOT:PSS electrodeposition; respect −0.6 V limit; design against delamination | §43 | Order-of-magnitude impedance win on cheap electrodes |
| 9 | 60-electrode TiN MEA as standard consumable (€306, reusable) | §39 | Every protocol assumes this format |
| 10 | Adaptive reinforcement thresholds (cultures drift; fixed thresholds fail) | §34 | Difference between a demo and a working trainer |
| 11 | Neuron–astrocyte co-culture for viability | §37, §18 | Glia are infrastructure, not garnish |
| 12 | SpikeInterface + Kilosort4 + NWB as the software stack | — | Free, standard, publication-grade |
| 13 | Dehydration-preserved fungal components; humidity control = signal integrity | §29, §32 | Fungal parts are storable; humidity is an electrical spec |
| 14 | Mycelium as living multi-modal sensor | §31 | The near-term useful product of the fungal side |
| 15 | Bioprinting / defined-geometry placement of cells | §30, §48 | How you eventually place neurons and mycelium together |
| 16 | PEDOT:PSS-infused mycelium (make the fungus the electrode) | §54 | Highest-upside interface experiment |
| 17 | Decoder robustness over electrode count (Neuralink's software rescue) | §5 | Design decoders that survive electrode loss |
| 18 | Surface recording is “good enough” (Layer 7) | §8 | Validates the non-penetrating MEA approach |
| 19 | Peptide hydrogels for ≥5-month 3D culture | §45 | Long-term viability window for a living computer |
| 20 | Slime-mould network optimization as a design metaphor | §51–52 | Let the biology solve wiring problems |
| Raspberry Pi 5 4GB | $110 | [published] |
| ADS1256 24-bit ADC module | ~$13 | [reported] |
| Oyster grain spawn, 5 lb | $20 | [published] |
| Substrate + tub + humidity | ~$40 | [estimated] |
| Needle electrodes + wiring | ~$35 | [estimated] |
| Still-air box + agar supplies | ~$40 | [estimated] |
| microSD, power bank, cables | ~$40 | [estimated] |
Tier B — add human EEG: PiEEG-8 $350 [published] + electrodes/cap ~$100 → ~$450 more. First mycelium network alone (spawn, substrate, chamber, sterility): ~$100–250.
| Intan RHD2000 USB interface board | $2,755 | [published, older] |
| Intan RHD2132 32-ch headstage (or DIY $350–450) | $895 | [published, older] |
| MCS 60MEA TiN chips ×3 (€306 ea) | ~$1,000 | [published] |
| Primary rat cortical neurons ×1 vial | $822–908 | [published] |
| B-27 Plus culture system + poly-D-lysine | ~$460 | [published] |
| Incubator access (community lab, 3 mo) | $300 | [reported] |
| Consumables | ~$300 | [estimated] |
Tier D — own incubator instead: add $3,480–7,321 new / ~$4,750 used [published]. Media burn ≈ $300/month at list prices [estimated].
| Source | Price | Notes |
|---|---|---|
| SH-SY5Y neuroblastoma line (2 vials × 1M) | $395–420 | Cheapest; immortal; poor learning model — practise technique only |
| Primary rat cortical neurons, 4M (Gibco) | $908 | The right first real substrate; ~4–8 MEAs per vial; use on thaw, never re-freeze |
| Primary rat cortical neurons, ≥4M (Lonza) | $822 | Same class as above |
| iPSC-derived human neurons, 1M (bit.bio) | $1,069 list | $999/3-vial academic pack; the DishBrain/CL1 substrate class |
| iPSC-derived, other vendors (Ricoh) | $850–1,200 | Per vial |
| Neucyte SynFire glutamatergic, >7.5M | $1,550 | Per vial; MEA co-culture kit $3,675–4,600 |
| Neurobasal Medium 500 mL | $107–115 | Plus version $143 |
| B-27 Supplement 50X, 10 mL | $159 | Plus $194; full B-27 Plus system $303 |
| Poly-D-Lysine coating | $157 | All prices [published] |
Recommendation from the research: learn on SH-SY5Y, do first real experiments on primary rat cortical neurons, graduate to iPSC-derived only with funding. Vendors like ATCC require institutional biosafety sign-off before shipping mammalian lines — budget for affiliation, not just the vial.
MEA → Intan RHD/RHS (16-bit, 1–30 kS/s) → USB → Intan-RHX / Open Ephys GUI → SpikeInterface (bandpass 300–6,000 Hz, common-average ref) → Kilosort4 (GPU) or MountainSort5 (CPU) → Phy curation → NWB export → decoding / closed loop via RHS stim or LSL
Data reality: 32 ch × 30 kS/s ≈ 6.9 GB/h raw; 60-ch MEA ≈ 4.3 GB/h; HD-MEA ≈ 148 GB/h. Storage, not compute, is the bottleneck. Record to a big external drive; sort offline; keep raw only for key sessions.
Needle electrodes → ADS1256 / Pico ADC-24 (24-bit, 1 Hz) → CSV/HDF5 → spike-train segmentation + pattern clustering (§23) → slow features (drift, spike rate, inter-spike intervals) → LSL stream for fusion
Data reality: 1 Hz × 16 ch ≈ kilobytes/day. The fungal side is nearly free to store forever — log everything, continuously.
PiEEG-8 / OpenBCI → BrainFlow (BoardShim) → LSL → Python (MNE, scikit-learn) → P300 / SSVEP / motor-imagery classifiers
All tools free & open: SpikeInterface, Kilosort, MountainSort5, Phy, BrainFlow, Open Ephys, Intan-RHX, LSL, MNE-Python, Neo. Adopt NWB from day one — it's what makes data fundable and publishable. Neuronal loop latency: ms to tens of ms — adequate for §13/§33 protocols.
Strategy in one line: self-fund Phase 1 (~$300) → Experiment.com for Phases 2–3 (~$2,000–5,000) → community lab for Phase 4 → university partnership for Phases 5–7 and NIH-scale funding. Do not write an NIH grant before you have fungal data and a collaborator.
| Path | What it pays | What it requires | Realism |
|---|---|---|---|
| Experiment.com crowdfunding | Average raise ~$5,000 [reported]; 48.08% success rate | Clear, feasible, hypothesis-driven proposal; staff review; IRB for human/animal work; lab-notebook reporting. Fees 8% + 3–5% | Best fit for Phases 1–3. A fungal-BCI campaign is exactly the kind of vivid project that funds |
| Community lab (Genspace, BioCurious…) | Bench + incubator + TC room ~$100/mo (+$100 setup) | Membership, safety class, project safety-board review; BSL-1 only at BioCurious | The practical starting point for cell work — confirm per-lab what neuronal work is permitted |
| University partnership | Lab space, BSL-2, incubator, ATCC account, ethics approval, co-authored grants | A faculty collaborator; IBC protocol; shared IP terms | Highest-leverage move. Approach with Phase 1–2 fungal data in hand, not an idea — and the approach is stronger with Jake on the team: an accredited chemical engineer as wet-lab lead is exactly the credential a faculty collaborator wants to see |
| FinalSpark free access | Free Neuroplatform access for selected projects | A selected proposal | Worth applying with a sharp organoid-vs-mycelium comparison |
| NIH BRAIN Initiative | R01 tool-dev: $8M total / 6–9 awards; R61 up to $700k; SBIR Phase I→II | An institution — grants go to universities, hospitals, 501(c)(3)s, or small businesses, not individuals; PI + grants office + IBC | Only via partnership or a founded small business. 6–12 month lead. The endgame, not the start |
| Open-source sponsorship | GitHub Sponsors / Open Collective — amounts [unknown] | A public repo people use | Supplemental only — consumables money, not capital |
| Commercial rental (benchmarking) | CL1 cloud ~$300/wk; FinalSpark ~$500–1,000/mo [reported] | Payment; ~1 week provisioning lead (CL1) | Budget one month to benchmark DIY results — not a subscription |
Cheapest credible entry: the fungal rig, ~$200–300, digitized living signals in under a month. Highest learning per dollar after that: (1) replicate Adamatzky's recording + thermal-stimulus experiments — publishable replication, ~$300; (2) spike-train “language” statistics (§23) on your own data — free beyond the rig; (3) reservoir computing with mycelium as the reservoir (§15 architecture) — free beyond the rig, genuinely novel; (4) the full DishBrain protocol in simulation (Brian2/NEST) before touching cells — free.
Never skimp on: sterile-technique supplies (contamination is the #1 killer) · ADC resolution (fungal signals are ~0.5 mV — 24-bit is the floor, not a luxury) · electrode contact quality (beats channel count, §5) · incubator stability (a flaky DIY incubator is the most expensive “saving” in the project) · LN₂ storage (cells at −70 °C lose viability — don't order until you can store or plate immediately) · controls (medium-only, open-loop, unstimulated — in parallel, every time) · safety compliance.