geniml

Name: geniml
Author: LeonChaoX

698stars61forks

How to install geniml?

leonchaox/qinyan-academic-skills/geniml

$npx -y skills add LeonChaoX/qinyan-academic-skills --skill geniml

Installs into the current project.

Summary

This skill should be used when working with genomic interval data (BED files) for machine learning tasks. Use for training region embeddings (Region2Vec, BEDspace), single-cell ATAC-seq analysis (scEmbed), building consensus peaks (universes), or any ML-based analysis of genomic

Files · 1

View on GitHub

SKILL.md

1	# Geniml: Genomic Interval Machine Learning
2
3	## Overview
4
5	Geniml is a Python package for building machine learning models on genomic interval data from BED files. It provides unsupervised methods for learning embeddings of genomic regions, single cells, and metadata labels, enabling similarity searches, clustering, and downstream ML tasks.
6
7	## Installation
8
9	Install geniml using uv:
10
11	```bash
12	uv uv pip install geniml
13	```
14
15	For ML dependencies (PyTorch, etc.):
16
17	```bash
18	uv uv pip install 'geniml[ml]'
19	```
20
21	Development version from GitHub:
22
23	```bash
24	uv uv pip install git+https://github.com/databio/geniml.git
25	```
26
27	## Core Capabilities
28
29	Geniml provides five primary capabilities, each detailed in dedicated reference files:
30
31	### 1. Region2Vec: Genomic Region Embeddings
32
33	Train unsupervised embeddings of genomic regions using word2vec-style learning.
34
35	Use for: Dimensionality reduction of BED files, region similarity analysis, feature vectors for downstream ML.
36
37	Workflow:
38	1. Tokenize BED files using a universe reference
39	2. Train Region2Vec model on tokens
40	3. Generate embeddings for regions
41
42	Reference: See `references/region2vec.md` for detailed workflow, parameters, and examples.
43
44	### 2. BEDspace: Joint Region and Metadata Embeddings
45
46	Train shared embeddings for region sets and metadata labels using StarSpace.
47
48	Use for: Metadata-aware searches, cross-modal queries (region→label or label→region), joint analysis of genomic content and experimental conditions.
49
50	Workflow:
51	1. Preprocess regions and metadata
52	2. Train BEDspace model
53	3. Compute distances
54	4. Query across regions and labels
55
56	Reference: See `references/bedspace.md` for detailed workflow, search types, and examples.
57
58	### 3. scEmbed: Single-Cell Chromatin Accessibility Embeddings
59
60	Train Region2Vec models on single-cell ATAC-seq data for cell-level embeddings.
61
62	Use for: scATAC-seq clustering, cell-type annotation, dimensionality reduction of single cells, integration with scanpy workflows.
63
64	Workflow:
65	1. Prepare AnnData with peak coordinates
66	2. Pre-tokenize cells
67	3. Train scEmbed model
68	4. Generate cell embeddings
69	5. Cluster and visualize with scanpy
70
71	Reference: See `references/scembed.md` for detailed workflow, parameters, and examples.
72
73	### 4. Consensus Peaks: Universe Building
74
75	Build reference peak sets (universes) from BED file collections using multiple statistical methods.
76
77	Use for: Creating tokenization references, standardizing regions across datasets, defining consensus features with statistical rigor.
78
79	Workflow:
80	1. Combine BED files
81	2. Generate coverage tracks
82	3. Build universe using CC, CCF, ML, or HMM method
83
84	Methods:
85	- CC (Coverage Cutoff): Simple threshold-based
86	- CCF (Coverage Cutoff Flexible): Confidence intervals for boundaries
87	- ML (Maximum Likelihood): Probabilistic modeling of positions
88	- HMM (Hidden Markov Model): Complex state modeling
89
90	Reference: See `references/consensus_peaks.md` for method comparison, parameters, and examples.
91
92	### 5. Utilities: Supporting Tools
93
94	Additional tools for caching, randomization, evaluation, and search.
95
96	Available utilities:
97	- BBClient: BED file caching for repeated access
98	- BEDshift: Randomization preserving genomic context
99	- Evaluation: Metrics for embedding quality (silhouette, Davies-Bouldin, etc.)
100	- Tokenization: Region tokenization utilities (hard, soft, universe-based)
101	- Text2BedNN: Neural search backends for genomic queries
102
103	Reference: See `references/utilities.md` for detailed usage of each utility.
104
105	## Common Workflows
106
107	### Basic Region Embedding Pipeline
108
109	```python
110	from geniml.tokenization import hard_tokenization
111	from geniml.region2vec import region2vec
112	from geniml.evaluation import evaluate_embeddings
113
114	# Step 1: Tokenize BED files
115	hard_tokenization(
116	src_folder='bed_files/',
117	dst_folder='tokens/',
118	universe_file='universe.bed',
119	p_value_threshold=1e-9
120	)
121
122	# Step 2: Train Region2Vec
123	region2vec(
124	token_folder='tokens/',
125	save_dir='model/',
126	num_shufflings=1000,
127	embedding_dim=100
128	)
129
130	# Step 3: Evaluate
131	metrics = evaluate_embeddings(
132	embeddings_file='model/embeddings.npy',
133	labels_file='metadata.csv'
134	)
135	```
136
137	### scATAC-seq Analysis Pipeline
138
139	```python
140	import scanpy as sc
141	from geniml.scembed import ScEmbed
142	from geniml.io import tokenize_cells
143
144	# Step 1: Load data
145	adata = sc.read_h5ad('scatac_data.h5ad')
146
147	# Step 2: Tokenize cells
148	tokenize_cells(
149	adata='scatac_data.h5ad',

People who liked this also viewed

6 picks

Type

geniml

698stars61forks

View on GitHub

How to install geniml?

leonchaox/qinyan-academic-skills/geniml

$npx -y skills add LeonChaoX/qinyan-academic-skills --skill geniml

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Geniml: Genomic Interval Machine Learning
2
3	## Overview
4
5	Geniml is a Python package for building machine learning models on genomic interval data from BED files. It provides unsupervised methods for learning embeddings of genomic regions, single cells, and metadata labels, enabling similarity searches, clustering, and downstream ML tasks.
6
7	## Installation
8
9	Install geniml using uv:
10
11	```bash
12	uv uv pip install geniml
13	```
14
15	For ML dependencies (PyTorch, etc.):
16
17	```bash
18	uv uv pip install 'geniml[ml]'
19	```
20
21	Development version from GitHub:
22
23	```bash
24	uv uv pip install git+https://github.com/databio/geniml.git
25	```
26
27	## Core Capabilities
28
29	Geniml provides five primary capabilities, each detailed in dedicated reference files:
30
31	### 1. Region2Vec: Genomic Region Embeddings
32
33	Train unsupervised embeddings of genomic regions using word2vec-style learning.
34
35	Use for: Dimensionality reduction of BED files, region similarity analysis, feature vectors for downstream ML.
36
37	Workflow:
38	1. Tokenize BED files using a universe reference
39	2. Train Region2Vec model on tokens
40	3. Generate embeddings for regions
41
42	Reference: See `references/region2vec.md` for detailed workflow, parameters, and examples.
43
44	### 2. BEDspace: Joint Region and Metadata Embeddings
45
46	Train shared embeddings for region sets and metadata labels using StarSpace.
47
48	Use for: Metadata-aware searches, cross-modal queries (region→label or label→region), joint analysis of genomic content and experimental conditions.
49
50	Workflow:
51	1. Preprocess regions and metadata
52	2. Train BEDspace model
53	3. Compute distances
54	4. Query across regions and labels
55
56	Reference: See `references/bedspace.md` for detailed workflow, search types, and examples.
57
58	### 3. scEmbed: Single-Cell Chromatin Accessibility Embeddings
59
60	Train Region2Vec models on single-cell ATAC-seq data for cell-level embeddings.
61
62	Use for: scATAC-seq clustering, cell-type annotation, dimensionality reduction of single cells, integration with scanpy workflows.
63
64	Workflow:
65	1. Prepare AnnData with peak coordinates
66	2. Pre-tokenize cells
67	3. Train scEmbed model
68	4. Generate cell embeddings
69	5. Cluster and visualize with scanpy
70
71	Reference: See `references/scembed.md` for detailed workflow, parameters, and examples.
72
73	### 4. Consensus Peaks: Universe Building
74
75	Build reference peak sets (universes) from BED file collections using multiple statistical methods.
76
77	Use for: Creating tokenization references, standardizing regions across datasets, defining consensus features with statistical rigor.
78
79	Workflow:
80	1. Combine BED files
81	2. Generate coverage tracks
82	3. Build universe using CC, CCF, ML, or HMM method
83
84	Methods:
85	- CC (Coverage Cutoff): Simple threshold-based
86	- CCF (Coverage Cutoff Flexible): Confidence intervals for boundaries
87	- ML (Maximum Likelihood): Probabilistic modeling of positions
88	- HMM (Hidden Markov Model): Complex state modeling
89
90	Reference: See `references/consensus_peaks.md` for method comparison, parameters, and examples.
91
92	### 5. Utilities: Supporting Tools
93
94	Additional tools for caching, randomization, evaluation, and search.
95
96	Available utilities:
97	- BBClient: BED file caching for repeated access
98	- BEDshift: Randomization preserving genomic context
99	- Evaluation: Metrics for embedding quality (silhouette, Davies-Bouldin, etc.)
100	- Tokenization: Region tokenization utilities (hard, soft, universe-based)
101	- Text2BedNN: Neural search backends for genomic queries
102
103	Reference: See `references/utilities.md` for detailed usage of each utility.
104
105	## Common Workflows
106
107	### Basic Region Embedding Pipeline
108
109	```python
110	from geniml.tokenization import hard_tokenization
111	from geniml.region2vec import region2vec
112	from geniml.evaluation import evaluate_embeddings
113
114	# Step 1: Tokenize BED files
115	hard_tokenization(
116	src_folder='bed_files/',
117	dst_folder='tokens/',
118	universe_file='universe.bed',
119	p_value_threshold=1e-9
120	)
121
122	# Step 2: Train Region2Vec
123	region2vec(
124	token_folder='tokens/',
125	save_dir='model/',
126	num_shufflings=1000,
127	embedding_dim=100
128	)
129
130	# Step 3: Evaluate
131	metrics = evaluate_embeddings(
132	embeddings_file='model/embeddings.npy',
133	labels_file='metadata.csv'
134	)
135	```
136
137	### scATAC-seq Analysis Pipeline
138
139	```python
140	import scanpy as sc
141	from geniml.scembed import ScEmbed
142	from geniml.io import tokenize_cells
143
144	# Step 1: Load data
145	adata = sc.read_h5ad('scatac_data.h5ad')
146
147	# Step 2: Tokenize cells
148	tokenize_cells(
149	adata='scatac_data.h5ad',

People who liked this also viewed

6 picks

Type