🧬 Sachs Protein Signaling

Overview

Single-cell flow cytometry measurements of 11 phosphoproteins across 5,400 T-cells under 9 experimental conditions, from the landmark causal inference benchmark by Sachs et al. (2005).

Data Structure

Metric	Value
Proteins	Raf, Mek, Erk, Akt, PKA, PKC, P38, Jnk, Plcg, PIP2, PIP3
Cells	5,400 (1,800 observational + 3,600 interventional)
Conditions	6 (1 observational + 5 interventional reagents)
Ground-Truth DAG	18 signed causal edges (17 Sachs consensus arcs + PIP3 to Akt)
Citations	Sachs et al., Science 308:523-529 (2005); Guo & Dong, Linear Causal Discovery with Interventional Constraints (2025)

Causal Reasoning Model

The engine uses a signed ground-truth graph: each edge is either activating (high parent drives high child) or inhibitory (high parent drives low child). Topology follows Sachs et al. (2005) Table 3; edge signs follow the literature-curated interactions in Guo & Dong (2025) Table 5.

Three signaling pathways comprise the network:

1. MAPK Cascade: PKC activates Raf and Mek; Raf activates Mek; Mek activates Erk. 2. PKA hub (mostly inhibitory): PKA inhibits Raf (Ser259/14-3-3), Mek (net, via Raf), P38, Akt, and Jnk. Only PKA to Erk is modelled as activating. 3. Lipid signaling: Plcg to PIP3/PIP2; PIP3 to PIP2; PIP3 activates Akt.

Edge Provenance

Edge	Sign	Source
PKA to Raf	inhibits	Häfner 1994; Dumaz & Marais 2003 (Ser259 / 14-3-3)
PKA to Mek	inhibits	Net effect via Raf, Sachs 2005 Table 3 (indirect arc)
PKA to P38	inhibits	Metz et al. 2021
PKA to Akt	inhibits	Intervention arm (Akt LOW in 489/600 PKA cells); not yet literature-curated
PKA to Jnk	inhibits	Intervention arm (Jnk LOW in 600/600 PKA cells); not yet literature-curated
PKC to Mek	activates	Sachs 2005 Table 3 (via Raf S497/S499)
PIP3 to Akt	activates	Reactome R-HSA-1257604 (well established; not recovered by Sachs' algorithm)
all others	activate	Sachs 2005 consensus / Guo & Dong 2025 Table 5

Note: PKA to Akt and PKA to Jnk signs are data-derived from the intervention arm, not from a curated literature source (Sachs lists both as indirect arcs without a sign; Guo & Dong do not include them). They are flagged here for domain confirmation, though the interventional evidence is unambiguous.

Benchmark Significance

This is the most widely cited benchmark dataset in causal discovery. The logic engine reasons over the signed DAG and detects cells whose protein levels contradict the expected signed direction. Under PKA activation, low Raf/Mek/P38/Akt/Jnk is the expected cascade shutdown, so only ~22% of PKA cells are anomalous: those where a target escaped suppression (e.g. Raf or P38 stays HIGH) or where Erk unexpectedly dropped.

Sample Questions

• "Why does this cell show high Erk despite Mek inhibition?"
• "What causes signaling anomalies under PKA activation?"
• "Which cells violate the MAPK cascade?"