glmp / docs /paper /ProcessDSL_FlowCell10_Proposal_with_Mermaid.md
garywelz's picture
Upload 3 files
e6e36e2 verified
# ProcessDSL + FlowCell-10 Proposal
This proposal outlines a pilot initiative to integrate the **"genome as program"** concept and **cellular process flowcharting** into the Virtual Cell project.
The goal is to formalize biological processes as executable, interpretable programs that can be learned, simulated, and manipulated by AI.
---
## 1. ProcessDSL Specification
**ProcessDSL** is a domain-specific language for representing cellular processes.
It compiles human-readable flowcharts into machine-executable forms such as stochastic rule systems, Petri nets, or hybrid ODE/event simulators.
**Key features:**
- Reactions as rules with explicit guards and rate constants.
- Conditional logic (`IF/ELSE`) for regulation.
- Iterative loops (`WHILE`) for cyclic processes.
- Event triggers for environmental or signaling changes.
- Support for compartments (nucleus, cytosol, organelles).
---
## 2. FlowCell-10 Pilot Dataset
**FlowCell-10** is a curated set of ten well-characterized yeast pathways, each represented as:
1. A canonical flowchart
2. A ProcessDSL file
3. Reference simulation outputs from literature data
**Example pathways:**
1. Glycolysis
2. TOR nutrient sensing pathway
3. Heat shock response
4. Autophagy initiation
5. Unfolded protein response (UPR)
6. Cell cycle G1/S transition
7. Mitochondrial respiration control
8. Amino acid biosynthesis regulation
9. Gluconeogenesis
10. Alcoholic fermentation
---
## 3. Example ProcessDSL (Glycolysis)
```text
process Glycolysis in Cytosol:
state: [Glucose, G6P, F6P, F16BP, G3P, DHAP, PEP, Pyruvate, ATP, ADP, NAD+, NADH]
rule Hexokinase: Glucose + ATP -> G6P + ADP [guard: ATP>θ1]
rule PFK: F6P + ATP -> F16BP + ADP [guard: ATP<θ2 & AMP>θ3]
rule Aldolase: F16BP -> G3P + DHAP
rule TPI: DHAP <-> G3P
rule PyruvateKinase: PEP + ADP -> Pyruvate + ATP [allosteric: F16BP activates]
event GlucosePulse(t=0..T): inflow rate r_in
```
---
## 4. Expanded Glycolysis Flowchart (Mermaid)
Below is an example from FlowCell-10 showing **Glycolysis in Yeast** with branch and loop structure, rendered with Mermaid:
```mermaid
flowchart TD
A[Glucose Uptake<br/>(Transport into cell)]
--> B[Hexokinase<br/>Glucose → G6P]
B --> C[Isomerase<br/>G6P → F6P]
C --> D[Phosphofructokinase (PFK)<br/>F6P → F1,6BP]
%% Branch
D --> E1[DHAP<br/>(Dihydroxyacetone phosphate)]
D --> E2[G3P<br/>(Glyceraldehyde‑3‑phosphate)]
E1 -- TPI forward --> E2
E2 -- TPI reverse --> E1
%% Payoff phase
E2 --> F[G3P Oxidation & Phosphorylation<br/>(NADH + ATP yield)]
F --> G[Phosphoglycerate Mutase & Enolase<br/>→ PEP]
G --> H[Pyruvate Kinase<br/>PEP → Pyruvate + ATP]
H --> I[End Product:<br/>2 Pyruvate Molecules]
```
---
## 5. Deliverables
- ProcessDSL specification and parser.
- FlowCell-10 diagrams, DSL files, and simulation benchmarks.
- Jupyter notebook demo: diagram → ProcessDSL → simulation → data comparison.
- Documentation for extending the dataset.
---
## 6. Benefits to the Virtual Cell Project
- Provides an interpretable, executable representation of cellular processes.
- Bridges molecular prediction tools (e.g., AlphaFold 3) to systems-level dynamics.
- Enables counterfactual simulations and intervention planning.
- Creates training data for AI models to learn biological program induction.
---
## 7. Suggested DeepMind Contacts
1. **Demis Hassabis** – CEO, DeepMind (vision for Virtual Cell)
2. **Pushmeet Kohli** – Head of AI for Science, DeepMind
3. **John Jumper** – Lead researcher on AlphaFold
4. **Kathryn Tunyasuvunakool** – Research scientist, AlphaFold/biology modeling
5. **Alexander Zisserman** – Research scientist, graph and vision integration