Skip to content

Commit dae1012

Browse files
committed
Add 3DGS genetics 4x4 fanout plan
1 parent 9bfb371 commit dae1012

1 file changed

Lines changed: 131 additions & 0 deletions

File tree

Lines changed: 131 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,131 @@
1+
# 3DGS Genetics 4x4 Fanout Plan — lance-graph
2+
3+
## Goal
4+
5+
Explore how the 3DGS raw-field and 4x4 cognitive-shader representation can fan out into genetics and molecular sequence analysis.
6+
7+
This is an exploratory plan. It should not block the geospatial or ultrasound paths.
8+
9+
## Core analogy
10+
11+
```text
12+
3DGS geospatial field
13+
local anisotropic kernels over space
14+
15+
genetic field
16+
local anisotropic kernels over sequence, expression, structure, and time
17+
```
18+
19+
A genetic block is not literally a visual splat. It is a compact kernel carrying local evidence and uncertainty.
20+
21+
## 4x4 carrier interpretation
22+
23+
```text
24+
lane0: sequence coordinate / genomic locus / k-mer index
25+
lane1: transition state / motif covariance / edit-distance neighborhood
26+
lane2: expression / methylation / abundance / confidence
27+
lane3: time / sample / lineage / provenance
28+
```
29+
30+
## (4x4)^4 grammar
31+
32+
Use `(4x4)^4` as a four-level block hierarchy:
33+
34+
```text
35+
level 0: local motif carrier
36+
level 1: motif block
37+
level 2: gene/regulatory region block
38+
level 3: pathway / cell-state / population block
39+
```
40+
41+
This mirrors:
42+
43+
```text
44+
splat -> block -> tile -> region
45+
motif -> gene -> pathway -> phenotype
46+
```
47+
48+
## Data sources
49+
50+
Potential future sources:
51+
52+
```text
53+
FASTA / FASTQ
54+
variant calls
55+
expression matrices
56+
single-cell count matrices
57+
methylation tracks
58+
protein/domain annotations
59+
pathway graphs
60+
```
61+
62+
## Graph model
63+
64+
```text
65+
Sample -> SequenceFrame -> MotifBlock -> Gene -> Pathway -> Phenotype
66+
Sample -> ExpressionFrame -> ExpressionBlock -> CellType -> State
67+
```
68+
69+
## Tables
70+
71+
Possible tables:
72+
73+
```text
74+
genetic_sources
75+
genetic_frames
76+
motif_blocks
77+
gene_features
78+
variant_features
79+
expression_blocks
80+
genetic_certificates
81+
```
82+
83+
## Kernel-fit idea
84+
85+
Convert local sequence/expression regions into kernels:
86+
87+
```text
88+
sequence window
89+
-> k-mer / edit neighborhood / motif statistics
90+
-> 4x4 carrier
91+
-> certificate
92+
-> graph feature
93+
```
94+
95+
## Certificates
96+
97+
Possible certificate types:
98+
99+
```text
100+
motif sampling confidence
101+
variant neighborhood stability
102+
expression block variance
103+
batch-effect residual
104+
lineage transition confidence
105+
quantization error
106+
```
107+
108+
## HHTL relevance
109+
110+
Use HHTL-style cascade to avoid full dense scans:
111+
112+
```text
113+
HEEL: k-mer / hash prefix candidate
114+
HIP: edit-distance or motif score
115+
TWIG: expression/covariance refinement
116+
LEAF: exact alignment/statistical test
117+
```
118+
119+
## Acceptance criteria for promotion
120+
121+
Promote this into implementation only when:
122+
123+
- a small public fixture is selected
124+
- a domain-specific decoder is scoped
125+
- the 4x4 carrier has a precise biological meaning
126+
- certificates are meaningful and not decorative
127+
- no medical/diagnostic claims are made
128+
129+
## Boundary
130+
131+
This plan is for representation and research tooling. It is not a clinical genomics product plan.

0 commit comments

Comments
 (0)