Upload scCAFM model assets

README.md

@@ -9,7 +9,7 @@ tags:
 scCAFM is a causality-aware foundation model designed for large-scale single-cell transcriptomic analysis. Unlike existing single-cell foundation models that mainly learn associative gene relationships or operate only at the dataset‐ or cell-type level, scCAFM enables cell-specific causal inference at atlas scale while simultaneously learning transferable gene and cell embeddings enriched with causal semantics. By jointly modeling gene regulatory structure and context-dependent embeddings, scCAFM provides a powerful foundation for studying heterogeneous cellular states, developmental trajectories, disease progression, and perturbation responses.
 <br/>
 <div align=center>
-<img src="Fig1.png" width="70%">
+<img src="Fig1.png" width="90%">
 </div>
 <br/>
 

models/cond_dict.json

@@ -8,15 +8,15 @@
     "token_index": 1
   },
   {
-    "cond_value": "10x 3' v2",
+    "cond_value": "human",
     "token_index": 2
   },
   {
-    "cond_value": "seq-well",
+    "cond_value": "breast",
     "token_index": 3
   },
   {
-    "cond_value": "10x 5' v2",
+    "cond_value": "normal",
     "token_index": 4
   },
   {
@@ -24,71 +24,71 @@
     "token_index": 5
   },
   {
-    "cond_value": "human",
+    "cond_value": "10x 3' v2",
     "token_index": 6
   },
   {
-    "cond_value": "blood",
+    "cond_value": "drop-seq",
     "token_index": 7
   },
   {
-    "cond_value": "umbilical cord blood",
+    "cond_value": "smart-seq2",
     "token_index": 8
   },
   {
-    "cond_value": "normal",
+    "cond_value": "sci-rna-seq",
     "token_index": 9
   },
   {
-    "cond_value": "drop-seq",
+    "cond_value": "lung",
     "token_index": 10
   },
   {
-    "cond_value": "brain",
+    "cond_value": "microwell-seq",
     "token_index": 11
   },
   {
-    "cond_value": "left frontal lobe",
+    "cond_value": "10x 5' v2",
     "token_index": 12
   },
   {
-    "cond_value": "left temporal lobe",
+    "cond_value": "10x 5' transcription profiling",
     "token_index": 13
   },
   {
-    "cond_value": "right temporal lobe",
+    "cond_value": "lingula of left lung",
     "token_index": 14
   },
   {
-    "cond_value": "forebrain",
+    "cond_value": "lower lobe of left lung",
     "token_index": 15
   },
   {
-    "cond_value": "right frontal lobe",
+    "cond_value": "upper lobe of left lung",
     "token_index": 16
   },
   {
-    "cond_value": "frontal lobe",
+    "cond_value": "10x 3' transcription profiling",
     "token_index": 17
   },
   {
-    "cond_value": "parietal lobe",
+    "cond_value": "blood",
     "token_index": 18
   },
   {
-    "cond_value": "temporal lobe",
+    "cond_value": "thalamic complex",
     "token_index": 19
   },
   {
-    "cond_value": "occipital lobe",
+    "cond_value": "hypothalamus",
     "token_index": 20
   },
   {
-    "cond_value": "glioblastoma",
+    "cond_value": "midbrain",
     "token_index": 21
   },
   {
-    "cond_value": "cerebral nuclei",
+    "cond_value": "pons",
     "token_index": 22
   },
   {
@@ -96,203 +96,203 @@
     "token_index": 23
   },
   {
-    "cond_value": "dorsolateral prefrontal cortex",
+    "cond_value": "myelencephalon",
     "token_index": 24
   },
   {
-    "cond_value": "thalamic complex",
+    "cond_value": "cerebral nuclei",
     "token_index": 25
   },
   {
-    "cond_value": "primary motor cortex",
+    "cond_value": "hippocampal formation",
     "token_index": 26
   },
   {
-    "cond_value": "hypothalamus",
+    "cond_value": "cerebellum",
     "token_index": 27
   },
   {
-    "cond_value": "hippocampal formation",
+    "cond_value": "interventricular septum",
     "token_index": 28
   },
   {
-    "cond_value": "midbrain",
+    "cond_value": "heart left ventricle",
     "token_index": 29
   },
   {
-    "cond_value": "pons",
+    "cond_value": "heart right ventricle",
     "token_index": 30
   },
   {
-    "cond_value": "myelencephalon",
+    "cond_value": "apex of heart",
     "token_index": 31
   },
   {
-    "cond_value": "cerebellum",
+    "cond_value": "cardiac atrium",
     "token_index": 32
   },
   {
-    "cond_value": "gexscope technology",
+    "cond_value": "cardiac ventricle",
     "token_index": 33
   },
   {
-    "cond_value": "indrop",
+    "cond_value": "coronary artery",
     "token_index": 34
   },
   {
-    "cond_value": "bd rhapsody whole transcriptome analysis",
+    "cond_value": "heart",
     "token_index": 35
   },
   {
-    "cond_value": "lung",
+    "cond_value": "temporal cortex",
     "token_index": 36
   },
   {
-    "cond_value": "squamous cell lung carcinoma",
+    "cond_value": "primary motor cortex",
     "token_index": 37
   },
   {
-    "cond_value": "lung adenocarcinoma",
+    "cond_value": "prefrontal cortex",
     "token_index": 38
   },
   {
-    "cond_value": "non-small cell lung carcinoma",
+    "cond_value": "parietal cortex",
     "token_index": 39
   },
   {
-    "cond_value": "middle temporal gyrus",
+    "cond_value": "primary somatosensory cortex",
     "token_index": 40
   },
   {
-    "cond_value": "substantia nigra pars compacta",
+    "cond_value": "primary visual cortex",
     "token_index": 41
   },
   {
-    "cond_value": "breast",
+    "cond_value": "dorsolateral prefrontal cortex",
     "token_index": 42
   },
   {
-    "cond_value": "10x 5' transcription profiling",
+    "cond_value": "strt-seq",
     "token_index": 43
   },
   {
-    "cond_value": "lung parenchyma",
+    "cond_value": "bd rhapsody whole transcriptome analysis",
     "token_index": 44
   },
   {
-    "cond_value": "lower lobe of left lung",
+    "cond_value": "brain",
     "token_index": 45
   },
   {
-    "cond_value": "upper lobe of left lung",
+    "cond_value": "lymph node",
     "token_index": 46
   },
   {
-    "cond_value": "brodmann (1909) area 4",
+    "cond_value": "pleural effusion",
     "token_index": 47
   },
   {
-    "cond_value": "prefrontal cortex",
+    "cond_value": "liver",
     "token_index": 48
   },
   {
-    "cond_value": "orbitofrontal cortex",
+    "cond_value": "adrenal gland",
     "token_index": 49
   },
   {
-    "cond_value": "anterior cingulate cortex",
+    "cond_value": "axilla",
     "token_index": 50
   },
   {
-    "cond_value": "primary visual cortex",
+    "cond_value": "bone spine",
     "token_index": 51
   },
   {
-    "cond_value": "microwell-seq",
+    "cond_value": "left temporal lobe",
     "token_index": 52
   },
   {
-    "cond_value": "smart-seq v4",
+    "cond_value": "right frontal lobe",
     "token_index": 53
   },
   {
-    "cond_value": "mouse",
+    "cond_value": "right parietal lobe",
     "token_index": 54
   },
   {
-    "cond_value": "brain meninx",
+    "cond_value": "right temporal lobe",
     "token_index": 55
   },
   {
-    "cond_value": "sci-rna-seq",
+    "cond_value": "forebrain",
     "token_index": 56
   },
   {
-    "cond_value": "telencephalon",
+    "cond_value": "left frontal lobe",
     "token_index": 57
   },
   {
-    "cond_value": "parietal cortex",
+    "cond_value": "temporoparietal junction",
     "token_index": 58
   },
   {
-    "cond_value": "visual cortex",
+    "cond_value": "left parietal lobe",
     "token_index": 59
   },
   {
-    "cond_value": "entorhinal cortex",
+    "cond_value": "lung adenocarcinoma",
     "token_index": 60
   },
   {
-    "cond_value": "primary somatosensory cortex",
+    "cond_value": "small cell lung carcinoma",
     "token_index": 61
   },
   {
-    "cond_value": "agranular insular cortex",
+    "cond_value": "glioblastoma",
     "token_index": 62
   },
   {
-    "cond_value": "subicular complex",
+    "cond_value": "smart-seq v4",
     "token_index": 63
   },
   {
-    "cond_value": "retrosplenial granular cortex",
+    "cond_value": "anterior cingulate cortex",
     "token_index": 64
   },
   {
-    "cond_value": "posterior parietal association areas",
+    "cond_value": "angular gyrus",
     "token_index": 65
   },
   {
-    "cond_value": "auditory cortex",
+    "cond_value": "primary auditory cortex",
     "token_index": 66
   },
   {
-    "cond_value": "temporal cortex",
+    "cond_value": "middle temporal gyrus",
     "token_index": 67
   },
   {
-    "cond_value": "medial orbital frontal cortex",
+    "cond_value": "anterior cingulate gyrus",
     "token_index": 68
   },
   {
-    "cond_value": "lateral visual area",
+    "cond_value": "mouse",
     "token_index": 69
   },
   {
-    "cond_value": "caudate nucleus",
+    "cond_value": "diencephalon",
     "token_index": 70
   },
   {
-    "cond_value": "putamen",
+    "cond_value": "parietal lobe",
     "token_index": 71
   },
   {
-    "cond_value": "primary auditory cortex",
+    "cond_value": "temporal lobe",
     "token_index": 72
   },
   {
-    "cond_value": "angular gyrus",
+    "cond_value": "frontal lobe",
     "token_index": 73
   },
   {
@@ -300,307 +300,307 @@
     "token_index": 74
   },
   {
-    "cond_value": "interventricular septum",
+    "cond_value": "left cardiac atrium",
     "token_index": 75
   },
   {
-    "cond_value": "apex of heart",
+    "cond_value": "10x 3' v1",
     "token_index": 76
   },
   {
-    "cond_value": "heart left ventricle",
+    "cond_value": "lung parenchyma",
     "token_index": 77
   },
   {
-    "cond_value": "heart right ventricle",
+    "cond_value": "anterior wall of left ventricle",
     "token_index": 78
   },
   {
-    "cond_value": "left cardiac atrium",
+    "cond_value": "cortex of kidney",
     "token_index": 79
   },
   {
-    "cond_value": "smart-seq",
+    "cond_value": "kidney",
     "token_index": 80
   },
   {
-    "cond_value": "smart-seq2",
+    "cond_value": "renal medulla",
     "token_index": 81
   },
   {
-    "cond_value": "diencephalon",
+    "cond_value": "renal pelvis",
     "token_index": 82
   },
   {
-    "cond_value": "brain gray matter",
+    "cond_value": "kidney blood vessel",
     "token_index": 83
   },
   {
-    "cond_value": "brain white matter",
+    "cond_value": "non-small cell lung carcinoma",
     "token_index": 84
   },
   {
-    "cond_value": "barrel cortex",
+    "cond_value": "squamous cell lung carcinoma",
     "token_index": 85
   },
   {
-    "cond_value": "cerebellar hemisphere",
+    "cond_value": "substantia nigra pars compacta",
     "token_index": 86
   },
   {
-    "cond_value": "frontal cortex",
+    "cond_value": "entorhinal cortex",
     "token_index": 87
   },
   {
-    "cond_value": "trudrop",
+    "cond_value": "occipital lobe",
     "token_index": 88
   },
   {
-    "cond_value": "liver",
+    "cond_value": "umbilical cord blood",
     "token_index": 89
   },
   {
-    "cond_value": "lower lobe of right lung",
+    "cond_value": "occipital cortex",
     "token_index": 90
   },
   {
-    "cond_value": "upper lobe of right lung",
+    "cond_value": "frontal cortex",
     "token_index": 91
   },
   {
-    "cond_value": "right lung",
+    "cond_value": "seq-well",
     "token_index": 92
   },
   {
-    "cond_value": "left lung",
+    "cond_value": "brain meninx",
     "token_index": 93
   },
   {
-    "cond_value": "middle lobe of right lung",
+    "cond_value": "scalebio single cell rna sequencing",
     "token_index": 94
   },
   {
-    "cond_value": "transverse colon",
+    "cond_value": "retrosplenial granular cortex",
     "token_index": 95
   },
   {
-    "cond_value": "sigmoid colon",
+    "cond_value": "visual cortex",
     "token_index": 96
   },
   {
-    "cond_value": "descending colon",
+    "cond_value": "agranular insular cortex",
     "token_index": 97
   },
   {
-    "cond_value": "ascending colon",
+    "cond_value": "auditory cortex",
     "token_index": 98
   },
   {
-    "cond_value": "hepatic flexure of colon",
+    "cond_value": "subicular complex",
     "token_index": 99
   },
   {
-    "cond_value": "hepatic cecum",
+    "cond_value": "medial orbital frontal cortex",
     "token_index": 100
   },
   {
-    "cond_value": "rectum",
+    "cond_value": "lateral visual area",
     "token_index": 101
   },
   {
-    "cond_value": "kidney",
+    "cond_value": "posterior parietal association areas",
     "token_index": 102
   },
   {
-    "cond_value": "lymph node",
+    "cond_value": "telencephalon",
     "token_index": 103
   },
   {
-    "cond_value": "blastoma",
+    "cond_value": "brodmann (1909) area 4",
     "token_index": 104
   },
   {
-    "cond_value": "colorectal cancer",
+    "cond_value": "trudrop",
     "token_index": 105
   },
   {
-    "cond_value": "tubular adenoma",
+    "cond_value": "lower lobe of right lung",
     "token_index": 106
   },
   {
-    "cond_value": "tubulovillous adenoma",
+    "cond_value": "upper lobe of right lung",
     "token_index": 107
   },
   {
-    "cond_value": "clear cell renal carcinoma",
+    "cond_value": "right lung",
     "token_index": 108
   },
   {
-    "cond_value": "renal papilla",
+    "cond_value": "left lung",
     "token_index": 109
   },
   {
-    "cond_value": "cortex of kidney",
+    "cond_value": "middle lobe of right lung",
     "token_index": 110
   },
   {
-    "cond_value": "renal medulla",
+    "cond_value": "transverse colon",
     "token_index": 111
   },
   {
-    "cond_value": "cingulate cortex",
+    "cond_value": "sigmoid colon",
     "token_index": 112
   },
   {
-    "cond_value": "split-seq",
+    "cond_value": "descending colon",
     "token_index": 113
   },
   {
-    "cond_value": "10x 3' transcription profiling",
+    "cond_value": "ascending colon",
     "token_index": 114
   },
   {
-    "cond_value": "strt-seq",
+    "cond_value": "hepatic flexure of colon",
     "token_index": 115
   },
   {
-    "cond_value": "pleural effusion",
+    "cond_value": "hepatic cecum",
     "token_index": 116
   },
   {
-    "cond_value": "adrenal gland",
+    "cond_value": "rectum",
     "token_index": 117
   },
   {
-    "cond_value": "axilla",
+    "cond_value": "blastoma",
     "token_index": 118
   },
   {
-    "cond_value": "bone spine",
+    "cond_value": "colorectal cancer",
     "token_index": 119
   },
   {
-    "cond_value": "right parietal lobe",
+    "cond_value": "tubular adenoma",
     "token_index": 120
   },
   {
-    "cond_value": "temporoparietal junction",
+    "cond_value": "tubulovillous adenoma",
     "token_index": 121
   },
   {
-    "cond_value": "left parietal lobe",
+    "cond_value": "clear cell renal carcinoma",
     "token_index": 122
   },
   {
-    "cond_value": "small cell lung carcinoma",
+    "cond_value": "cerebellar cortex",
     "token_index": 123
   },
   {
-    "cond_value": "body of stomach",
+    "cond_value": "hemisphere part of cerebellar posterior lobe",
     "token_index": 124
   },
   {
-    "cond_value": "gastric cancer",
+    "cond_value": "dentate nucleus",
     "token_index": 125
   },
   {
-    "cond_value": "lateral entorhinal cortex",
+    "cond_value": "split-seq",
     "token_index": 126
   },
   {
-    "cond_value": "retrosplenial region",
+    "cond_value": "caudate lobe of liver",
     "token_index": 127
   },
   {
-    "cond_value": "anterolateral visual area",
+    "cond_value": "bd rhapsody targeted mrna",
     "token_index": 128
   },
   {
-    "cond_value": "secondary somatosensory cortex",
+    "cond_value": "venous blood",
     "token_index": 129
   },
   {
-    "cond_value": "medial entorhinal cortex",
+    "cond_value": "gexscope technology",
     "token_index": 130
   },
   {
-    "cond_value": "gustatory cortex",
+    "cond_value": "adrenal tissue",
     "token_index": 131
   },
   {
-    "cond_value": "claustrum of brain",
+    "cond_value": "indrop",
     "token_index": 132
   },
   {
-    "cond_value": "caudate lobe of liver",
+    "cond_value": "basal ganglion",
     "token_index": 133
   },
   {
-    "cond_value": "occipital cortex",
+    "cond_value": "intestine",
     "token_index": 134
   },
   {
-    "cond_value": "renal pelvis",
+    "cond_value": "lateral entorhinal cortex",
     "token_index": 135
   },
   {
-    "cond_value": "kidney blood vessel",
+    "cond_value": "retrosplenial region",
     "token_index": 136
   },
   {
-    "cond_value": "heart",
+    "cond_value": "anterolateral visual area",
     "token_index": 137
   },
   {
-    "cond_value": "scalebio single cell rna sequencing",
+    "cond_value": "secondary somatosensory cortex",
     "token_index": 138
   },
   {
-    "cond_value": "10x 3' v1",
+    "cond_value": "medial entorhinal cortex",
     "token_index": 139
   },
   {
-    "cond_value": "adrenal tissue",
+    "cond_value": "gustatory cortex",
     "token_index": 140
   },
   {
-    "cond_value": "bone marrow",
+    "cond_value": "claustrum of brain",
     "token_index": 141
   },
   {
-    "cond_value": "b-cell non-hodgkin lymphoma",
+    "cond_value": "alveolus of lung",
     "token_index": 142
   },
   {
-    "cond_value": "follicular lymphoma",
+    "cond_value": "body of stomach",
     "token_index": 143
   },
   {
-    "cond_value": "anterior wall of left ventricle",
+    "cond_value": "gastric cancer",
     "token_index": 144
   },
   {
-    "cond_value": "cardiac atrium",
+    "cond_value": "cel-seq2",
     "token_index": 145
   },
   {
-    "cond_value": "cardiac ventricle",
+    "cond_value": "renal papilla",
     "token_index": 146
   },
   {
-    "cond_value": "coronary artery",
+    "cond_value": "upper outer quadrant of breast",
     "token_index": 147
   },
   {
-    "cond_value": "alveolus of lung",
+    "cond_value": "caudate nucleus",
     "token_index": 148
   },
   {
-    "cond_value": "superior frontal gyrus",
+    "cond_value": "putamen",
     "token_index": 149
   },
   {
-    "cond_value": "basal ganglion",
+    "cond_value": "cingulate cortex",
     "token_index": 150
   },
   {
@@ -608,103 +608,103 @@
     "token_index": 151
   },
   {
-    "cond_value": "dronc-seq",
+    "cond_value": "bone marrow",
     "token_index": 152
   },
   {
-    "cond_value": "mesonephros",
+    "cond_value": "b-cell non-hodgkin lymphoma",
     "token_index": 153
   },
   {
-    "cond_value": "renal glomerulus",
+    "cond_value": "follicular lymphoma",
     "token_index": 154
   },
   {
-    "cond_value": "outer medulla of kidney",
+    "cond_value": "smart-seq",
     "token_index": 155
   },
   {
-    "cond_value": "inner medulla of kidney",
+    "cond_value": "brain gray matter",
     "token_index": 156
   },
   {
-    "cond_value": "anterior cingulate gyrus",
+    "cond_value": "brain white matter",
     "token_index": 157
   },
   {
-    "cond_value": "cel-seq2",
+    "cond_value": "barrel cortex",
     "token_index": 158
   },
   {
-    "cond_value": "bd rhapsody targeted mrna",
+    "cond_value": "cerebellar hemisphere",
     "token_index": 159
   },
   {
-    "cond_value": "venous blood",
+    "cond_value": "ileum",
     "token_index": 160
   },
   {
-    "cond_value": "cerebellar cortex",
+    "cond_value": "acute myeloid leukemia",
     "token_index": 161
   },
   {
-    "cond_value": "hemisphere part of cerebellar posterior lobe",
+    "cond_value": "acute promyelocytic leukemia",
     "token_index": 162
   },
   {
-    "cond_value": "dentate nucleus",
+    "cond_value": "chromophobe renal cell carcinoma",
     "token_index": 163
   },
   {
-    "cond_value": "basal zone of heart",
+    "cond_value": "adenocarcinoma",
     "token_index": 164
   },
   {
-    "cond_value": "intestine",
+    "cond_value": "neuroendocrine carcinoma",
     "token_index": 165
   },
   {
-    "cond_value": "ileum",
+    "cond_value": "wilms tumor",
     "token_index": 166
   },
   {
-    "cond_value": "acute myeloid leukemia",
+    "cond_value": "pilocytic astrocytoma",
     "token_index": 167
   },
   {
-    "cond_value": "acute promyelocytic leukemia",
+    "cond_value": "orbitofrontal cortex",
     "token_index": 168
   },
   {
-    "cond_value": "chromophobe renal cell carcinoma",
+    "cond_value": "right occipital lobe",
     "token_index": 169
   },
   {
-    "cond_value": "adenocarcinoma",
+    "cond_value": "dronc-seq",
     "token_index": 170
   },
   {
-    "cond_value": "neuroendocrine carcinoma",
+    "cond_value": "mesonephros",
     "token_index": 171
   },
   {
-    "cond_value": "wilms tumor",
+    "cond_value": "renal glomerulus",
     "token_index": 172
   },
   {
-    "cond_value": "pilocytic astrocytoma",
+    "cond_value": "outer medulla of kidney",
     "token_index": 173
   },
   {
-    "cond_value": "upper outer quadrant of breast",
+    "cond_value": "inner medulla of kidney",
     "token_index": 174
   },
   {
-    "cond_value": "right occipital lobe",
+    "cond_value": "superior frontal gyrus",
     "token_index": 175
   },
   {
-    "cond_value": "lingula of left lung",
+    "cond_value": "basal zone of heart",
     "token_index": 176
   }
 ]

models/sfm_config.json

@@ -1,19 +1,18 @@
 {
   "sfm": {
     "embed_dim": 768,
-    "expr_hidden_dim": 768,
-    "expr_dropout": 0.1,
-    "expr_value_scale": 1.0,
-    "mod_hidden_dim": 768,
-    "mod_dropout": 0.1,
+    "expr_num_bins": 32,
+    "expr_hidden_dim": 128,
+    "expr_tau": 1.0,
+    "batch_num_bins": 128,
+    "batch_hidden_dim": 128,
+    "batch_tau": 1.0,
     "cond_dropout": 0.1,
-    "context_hidden_dim": 768,
-    "context_dropout": 0.1,
-    "embedding_dropout": 0.1,
+    "out_dropout": 0.1,
     "freeze_loaded_gene_embeddings": false,
     "num_layers": 12,
     "num_heads": 12,
-    "backbone_mlp_hidden_dim": 2048,
+    "backbone_mlp_hidden_dim": 3072,
     "backbone_mlp_dropout": 0.02,
     "use_rotary": false,
     "qkv_bias": true,
@@ -25,7 +24,7 @@
     "num_factors": 256,
     "router_hidden_dim": 1024,
     "router_dropout": 0.1,
-    "topk": 24,
+    "topk": 32,
     "router_temperature": 0.7,
     "beta_momentum": 0.1
   },

models/sfm_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:748070c393b74de89998967aa6fe79bfda4e98db43184072e23b86f29e0dcb81
-size 605872652
+oid sha256:e010e29145bbd3b2bce5aca9d188cfd5aa6aeae32dcfd5534ea65f67967c95c5
+size 541486920