Fungus vs Saline Spread Comparison

Summary Statistics

Fungus Window (Nov 6-12, 2025)

ESP	N (5-min)	Mean	P95	P99	Max	>150	>300	% >150	% >300
ESP01	1,644	33.2	61.2	84.9	197.8	2	0	0.1%	0.0%
ESP05	1,644	49.5	234.4	272.7	322.6	165	2	10.0%	0.1%
ESP08	1,644	15.6	28.3	41.8	89.8	0	0	0.0%	0.0%
ESP11	1,643	30.6	50.7	61.9	81.6	0	0	0.0%	0.0%
ESP13	1,644	35.1	55.6	68.4	90.8	0	0	0.0%	0.0%
ESP14	1,644	20.0	37.3	50.6	162.6	1	0	0.1%	0.0%
ESP16	1,644	268.2	1925.8	2575.3	2984.2	298	277	18.1%	16.8%
ESP17	1,644	54.3	134.5	158.9	170.3	28	0	1.7%	0.0%

Saline Window (Oct 3-9, 2025)

ESP	N (5-min)	Mean	P95	P99	Max	>150	>300	% >150	% >300
ESP04	1,828	385.8	761.8	890.7	1263.8	1,689	1,125	92.4%	61.5%
ESP05	1,977	524.5	920.1	1030.3	1187.4	1,916	1,715	96.9%	86.7%
ESP06	1,976	101.5	251.1	378.6	718.2	443	48	22.4%	2.4%
ESP07	1,977	332.6	1162.3	1544.5	1734.3	1,174	564	59.4%	28.5%
ESP08	1,977	73.4	180.8	265.0	405.3	265	6	13.4%	0.3%
ESP09	1,977	171.5	400.9	735.2	1003.9	894	198	45.2%	10.0%
ESP11	1,828	224.6	500.0	587.9	624.9	1,075	610	58.8%	33.4%
ESP13	1,828	269.0	852.3	1041.1	1309.4	777	660	42.5%	36.1%
ESP14	1,828	303.9	963.6	1135.9	1339.3	1,036	643	56.7%	35.2%
ESP17	1,828	140.7	327.9	452.2	1397.6	722	131	39.5%	7.2%

Final Verdict

Are high-spread episodes in living fungus indistinguishable from saline evaporation/heat artifacts?

NO — Most fungus spreads are significantly lower than saline, but ESP16 shows extreme outliers (max 2984 counts) that exceed saline late period. This suggests: (1) typical fungal spreads are biological and much tighter than environmental artifacts; (2) ESP16's extreme episodes are a special case requiring separate investigation.

Evidence

Mean spread comparison:

Fungus (Nov 6-12): 63.31 counts (median: 26.64)
Saline Early (Oct 3-6): 156.95 counts (median: 111.31)
Saline Late (Oct 7-9): 375.34 counts (median: 292.53)

Maximum spread:

Fungus overall: 2984.24 counts (driven by ESP16: 2984.24)
Fungus excluding ESP16: 322.61 counts
Saline Early: 852.47 counts
Saline Late: 1734.31 counts

ESP16 outlier analysis:

ESP16 mean spread: 268.2 counts (vs overall fungus mean: 63.3)
ESP16 max spread: 2984.2 counts (exceeds all saline maxes)
ESP16 % above 150: 18.1% (vs fungus overall: 3.8%)
ESP16 % above 300: 16.8% (vs fungus overall: 2.1%)
Note: ESP16's extreme values (documented in 001B addendum) are likely related to a2 electrode behavior, not general fungal spread patterns.

High-spread episodes (>150 counts):

Fungus: 3.8% of timestamps
Saline Early: 38.5% of timestamps
Saline Late: 70.7% of timestamps

Statistical tests (Mann-Whitney U):

Fungus vs Saline Early: p=0.00e+00 (significantly different)
Fungus vs Saline Late: p=0.00e+00 (significantly different)
Saline Early vs Late: p=0.00e+00 (significantly different - confirms environmental change)

Assessment of 001B report claim:

The 001B report states: 'The transient increases in fungus (like ESP05's 176-count episode) are comparable to saline's late-day spreads, suggesting both reflect interface/environmental shifts rather than biological state changes.'

REFUTES: Fungus spreads are significantly different from saline late period (p=0.00e+00), suggesting different mechanisms or additional biological contributions.

Methodology

Why This Analysis Was Critical

Experiment 001B documented electrode spread patterns (standard deviation across a0–a3) in the living fungus baseline. The 001B report stated that "transient increases in fungus (like ESP05's 176-count episode) are comparable to saline's late-day spreads, suggesting both reflect interface/environmental shifts rather than biological state changes." This claim needed rigorous quantitative validation:

Key question: Are high-spread episodes in living fungus statistically and visually indistinguishable from saline evaporation/heat artifacts, or are they clearly different biological phenomena?
Implication: If spreads are identical to saline artifacts, the fungal electrical activity may be driven primarily by environmental factors (temperature, evaporation, interface drift) rather than biological processes. If they are significantly different, this suggests biological mechanisms contribute to electrode spread patterns.
Importance: This distinction is fundamental to interpreting whether observed electrical patterns represent fungal physiology or hardware/environmental artifacts. Without this validation, we cannot confidently attribute electrical signatures to biological processes.

Exact Method Used

Step 1: Data Extraction

Fungus dataset: Loaded per-ESP CSV files from Experiment 001A window (Nov 6–12, 2025). Files: esp01_experiment_001a.csv through esp17_experiment_001a.csv (8 ESPs total: esp01, esp05, esp08, esp11, esp13, esp14, esp16, esp17).
Saline dataset: Loaded per-ESP CSV files from saline per-electrode extraction (Oct 3–9, 2025). Files: esp04_saline_per_electrode.csv through esp17_saline_per_electrode.csv (10 ESPs total).
Saline period splitting: Divided saline period into early (Oct 3–6) and late (Oct 7–9) to account for known warming/evaporation effects starting around Oct 6–7.

Step 2: Resampling (Matching 001B Method)

Each channel (a0, a1, a2, a3) was resampled to 5-minute median values using pandas resample("5min").median().
Why median (not mean): Median is robust to outliers and matches the exact method used in Experiment 001B, ensuring consistency across analyses.
Why 5 minutes: This time window smooths high-frequency noise while preserving diurnal patterns and transient events. It matches the standard 001B processing pipeline.

Step 3: Spread Calculation

At each 5-minute timestamp, computed spread = std([a0, a1, a2, a3]) using pandas std(axis=1) across the four electrode channels.
Why standard deviation: This metric quantifies how tightly clustered the four electrodes are at each moment. Low spread (e.g., <50 counts) indicates electrodes are sampling a homogeneous electrical domain. High spread (e.g., >150 counts) indicates electrode separation or differential activity.
This produces one spread value per 5-minute timestamp per ESP.

Step 4: Summary Statistics

For each ESP in each dataset, computed: mean spread, median spread, P95, P99, maximum spread, and counts/t percentages above thresholds (150 and 300 counts).
Why multiple metrics: Mean captures central tendency, but median is more robust to outliers (especially important given ESP16's extreme values). Percentiles capture distribution tails. Threshold counts identify high-spread episodes.

Step 5: Statistical Comparison

Performed Mann-Whitney U tests (non-parametric) to compare distributions: fungus vs saline early, fungus vs saline late, and saline early vs late.
Why Mann-Whitney U: Non-parametric test that doesn't assume normal distributions. Appropriate for comparing spread distributions which may be skewed or have heavy tails.
Why not t-test: Spread distributions are likely non-normal (especially with ESP16 outliers), so parametric tests would be inappropriate.

Step 6: Temperature Correlation (Saline Period)

Attempted to load weather data (/data/weather_oct2025.parquet) to compute Pearson correlation between daily max temperature and daily mean spread for saline period.
Why: If spreads correlate strongly with temperature in saline, this validates that saline spreads are environmental artifacts. The absence of temperature data doesn't invalidate the comparison, as the saline early vs late split already captures the warming period.

Data Windows

Fungus window: 2025-11-06 16:01 UTC → 2025-11-12 15:00 UTC (identical to Experiment 001A/001B baseline)
Saline window: 2025-10-03 00:00 UTC → 2025-10-10 00:00 UTC (full saline control period)
Saline early period: Oct 3–6, 2025 (before known warming/evaporation effects)
Saline late period: Oct 7–9, 2025 (during known warming/evaporation period)

Script

Analysis performed using: analysis/fungal_state_mapping/scripts/compare_fungus_vs_saline_spread.py

Fungus vs Saline Spread Comparison Analysis (Analyzed by AI)

Visualizations