Predicting the effects of genetic perturbations requires models that generalize beyond observed cell contexts and gene combinations. We introduce Occam, a multilayer perceptron (MLP) framework for representation learning and perturbation-effect prediction. Occam combines cell representations learned through denoising with gene interaction, text, and protein-sequence features. It achieves the highest aggregate representation scores for perturbation information and cell-type preservation with batch integration. In cross-context few-shot prediction, a perturbation and a cell context are each observed during training, but never together. Occam better captures the measured response patterns for these unobserved combinations than four retrained competitors or direct transfer of the same perturbation's measured effects from other training cell lines. Without fine-tuning, Occam best distinguishes perturbations among the evaluated models and simple baselines in cell lines held out from predictor training, across seen and unseen assays. Predicting expression levels accurately in these settings remains challenging. For double perturbations, which target two genes simultaneously, Occam composes learned single-gene responses to better capture the combined response and reduce expression error compared with competing models and summed single-gene effects. These gains extend to both strongly additive and weakly additive responses. Together, these results support simple MLPs for cell representation learning and perturbation-effect prediction across contexts and gene combinations.
Figures & tables
Figure 1: Occam ’s training stages. Left, representation learning: reconstructing expression and detecting entries replaced from one donor cell train the encoder. The negative-binomial (NB) loss follows expm1 , which inverts the log transform. Right, perturbation-effect prediction: a frozen cell encoder, perturbation multilayer perceptron (MLP), modulation module, and trainable decoder predict perturbed expression from a control cell. Perturbation features describe gene networks (GO+STRING), gene text (GenePT), and protein sequences (ESM2). SmoothL1 supervises expression levels; a cosine loss aligns changes relative to controls with the observed mean effect.
Perturbation representation ↑
scIB ↑
Model
RPE1
HepG2
Jurkat
Overall
Pancreas
PBMC-68k
Overall
Nonlinear encoders Trained on DiverseRNA ( 1.4M profiles )
Occam-90M
42.94 ± 0.04
38.79 ± 0.04
36.22 ± 0.02
39.31 ± 0.02
53.10 ± 0.12
54.33 ± 0.16
53.72 ± 0.10
TxFM-B
42.05 ± 0.02
37.50 ± 0.02
36.81 ± 0.01
38.78 ± 0.01
47.29 ± 0.10
53.14 ± 0.24
50.22 ± 0.13
scVI-110M
35.55 ± 0.08
33.29 ± 0.08
30.62 ± 0.04
33.16 ± 0.04
48.72 ± 0.18
55.02 ± 0.24
51.87 ± 0.15
STATE-SE-90M
35.86 ± 0.20
32.14 ± 0.20
30.78 ± 0.17
32.93 ± 0.11
47.03 ± 0.30
48.38 ± 0.16
47.71 ± 0.17
Table 1: Cell-representation quality by dataset and overall. Trained methods in the first two groups share the DiverseRNA corpus; public checkpoints use much larger corpora. Bold and underline mark the best and second-best mean, respectively, in each column across all groups.
Figure 2: Perturbation-effect prediction across cell lines. RHJ-trained models are evaluated on unseen perturbation–HepG2 combinations (few-shot), K562 (seen assay), and H1 (unseen assay), without fine-tuning. Transfer uses the same perturbation’s measured effects from other training cell lines. Points show five training-seed means over five evaluation splits each. Boxes show quartiles and medians, whiskers show minima and maxima, and diamonds and labels show means. Dotted lines show simple-baseline means; dash-dot lines show experimental reproducibility (not a ceiling). Linear ranges vary by panel. MAE values are scaled by 103 . Means and SDs: Tables 2 – 4 .
Figure 3: Perturbation-effect prediction for held-out double perturbations. Norman’s 111 test double perturbations include 36 strongly additive, 36 weakly additive, and 39 intermediate responses. Points show five training-seed means over five evaluation splits each. Boxes show quartiles and medians, whiskers show minima and maxima, and diamonds and labels show means. Dotted lines show simple-baseline means; dash-dot lines show experimental reproducibility (not a ceiling). Linear ranges vary by panel. MAE is scaled by 103 . Means and SDs: Tables 6 , 7 , and 9 .
Appendix figures & tables22 assets
Supplementary material from the paper’s appendix.
Appendix
Model
Pearson Δ↑
MAE ↓
Overlap@N ↑
L1 discrimination ↑
Occam
0.3999±0.0035
0.02688±0.00203
0.0860±0.0066
0.7610±0.0018
TxPert
0.3438±0.0092
0.02278±0.00032
0.0557±0.0064
0.6658±0.0090
STATE
0.2492±0.0205
0.02769±0.00134
0.0887±0.0080
0.6684±0.0164
CellFlow
0.2135±0.0084
0.02495±0.00063
0.0316±0.0034
0.5467±0.0098
PerturbDiff
0.2491±0.0080
0.02648±0.00123
0.0644±0.0063
0.6435±0.0152
Transfer
0.2332±0.0005
0.02986±0.00001
0.0353±0.0004
0.6416±0.0010
Appendix
Table 2: Perturbation-effect prediction: RHJ/HepG2, cross-context few-shot prediction (611 perturbations). Values are mean ± SD across five seed means, each averaged over five evaluation splits (Appendix E.4 ).
Model
Pearson Δ↑
MAE ↓
Overlap@N ↑
L1 discrimination ↑
Occam
0.1071±0.0057
0.14232±0.00121
0.3250±0.0012
0.6436±0.0132
TxPert
0.0868±0.0015
0.14895±0.00448
0.3385±0.0069
0.5155±0.0094
STATE
0.0473±0.0102
0.19963±0.02696
0.3188±0.0126
0.5073±0.0044
CellFlow
0.0445±0.0114
0.17320±0.01123
0.2826±0.0089
0.5044±0.0032
PerturbDiff
0.0879±0.0025
0.14545±0.00247
0.3257±0.0015
0.5237±0.0011
Transfer
0.2324±0.0012
0.01749±0.00001
0.2542±0.0003
0.5695±0.0003
Appendix
Table 3: Perturbation-effect prediction: H1, unseen cell line and unseen assay (80 perturbations). Values are mean ± SD across five seed means, each averaged over five evaluation splits (Appendix E.4 ).
Model
Pearson Δ↑
MAE ↓
Overlap@N ↑
L1 discrimination ↑
Occam
0.2115±0.0125
0.05619±0.00094
0.0639±0.0031
0.7488±0.0029
TxPert
0.1446±0.0048
0.04742±0.00025
0.0475±0.0017
0.6110±0.0055
STATE
0.1341±0.0138
0.05368±0.00232
0.0348±0.0060
0.5998±0.0183
CellFlow
0.1167±0.0067
0.06171±0.00285
0.0355±0.0017
0.5212±0.0079
PerturbDiff
0.1232±0.0076
0.05964±0.00319
0.0390±0.0008
0.5642±0.0164
Transfer
0.2710±0.0003
0.02454±0.00001
0.0533±0.0004
0.7090±0.0008
Appendix
Table 4: Perturbation-effect prediction: K562, unseen cell line and seen assay (528 perturbations). Values are mean ± SD across five seed means, each averaged over five evaluation splits (Appendix E.4 ).
Model
Pearson Δ↑
MAE ↓
Overlap@N ↑
L1 discrimination ↑
Occam
0.2033±0.0138
0.05532±0.00082
0.0715±0.0045
0.6981±0.0048
TxPert
0.1323±0.0067
0.04672±0.00024
0.0467±0.0021
0.6071±0.0029
STATE
0.1143±0.0131
0.05221±0.00260
0.0331±0.0069
0.5466±0.0176
CellFlow
0.0857±0.0118
0.05916±0.00344
0.0315±0.0021
0.5211±0.0045
PerturbDiff
0.1101±0.0100
0.05841±0.00329
0.0368±0.0015
0.5413±0.0065
Transfer
0.2819±0.0012
0.02238±0.00003
0.0602±0.0014
0.6627±0.0018
Appendix
Table 5: Perturbation-effect prediction: K562, the same 80 target genes as H1. Values are mean ± SD across five seed means, each averaged over five evaluation splits (Appendix E.4 ).
Model
Pearson Δ↑
MAE ↓
Overlap@N ↑
L1 discrimination ↑
Occam
0.8955±0.0032
0.01079±0.00016
0.2843±0.0065
0.9662±0.0046
TxPert
0.8513±0.0080
0.01363±0.00019
0.3236±0.0021
0.9747±0.0031
STATE
0.6989±0.0385
0.01623±0.00083
0.2771±0.0277
0.9066±0.0143
CellFlow
0.7655±0.0154
0.01390±0.00012
0.2493±0.0083
0.9308±0.0153
PerturbDiff
0.6582±0.0960
0.01853±0.00285
0.1651±0.0207
0.7601±0.0415
Sum of singles
0.8822±0.0003
0.01262±0.00001
0.2864±0.0012
0.9890±0.0003
Appendix
Table 6: Perturbation-effect prediction: Norman, held-out double perturbations in K562. 36 strongly additive double perturbations, grouped by effect size. Values are mean ± SD across five seed means, each averaged over five evaluation splits (Appendix E.4 ).
Model
Pearson Δ↑
MAE ↓
Overlap@N ↑
L1 discrimination ↑
Occam
0.8013±0.0044
0.01339±0.00014
0.2720±0.0048
0.8249±0.0123
TxPert
0.7069±0.0069
0.01916±0.00024
0.3009±0.0020
0.7156±0.0083
STATE
0.6316±0.0303
0.01868±0.00101
0.2600±0.0180
0.7500±0.0114
CellFlow
0.6984±0.0168
0.01659±0.00040
0.2531±0.0090
0.7751±0.0086
PerturbDiff
0.6128±0.0627
0.01986±0.00203
0.1744±0.0213
0.7357±0.0133
Sum of singles
0.7749±0.0012
0.01679±0.00004
0.2751±0.0011
0.8567±0.0029
Appendix
Table 7: Perturbation-effect prediction: Norman, held-out double perturbations in K562. 36 weakly additive double perturbations, grouped by effect size. Values are mean ± SD across five seed means, each averaged over five evaluation splits (Appendix E.4 ).
Model
Pearson Δ↑
MAE ↓
Overlap@N ↑
L1 discrimination ↑
Occam
0.8604±0.0009
0.01254±0.00007
0.2852±0.0061
0.8975±0.0253
TxPert
0.7838±0.0085
0.01679±0.00020
0.3442±0.0025
0.8152±0.0110
STATE
0.6681±0.0348
0.01816±0.00067
0.2812±0.0260
0.7304±0.0140
CellFlow
0.7147±0.0287
0.01614±0.00022
0.2656±0.0101
0.7486±0.0309
PerturbDiff
0.6052±0.1023
0.02025±0.00259
0.1755±0.0222
0.6633±0.0320
Sum of singles
0.8423±0.0006
0.01455±0.00001
0.2875±0.0005
0.9409±0.0008
Appendix
Table 8: Perturbation-effect prediction: Norman, held-out double perturbations in K562. 39 intermediate double perturbations, grouped by effect size. Values are mean ± SD across five seed means, each averaged over five evaluation splits (Appendix E.4 ).
Model
Pearson Δ↑
MAE ↓
Overlap@N ↑
L1 discrimination ↑
Occam
0.8526±0.0007
0.01225±0.00004
0.2806±0.0055
0.8962±0.0117
TxPert
0.7808±0.0072
0.01653±0.00021
0.3235±0.0018
0.8346±0.0072
STATE
0.6663±0.0335
0.01770±0.00082
0.2730±0.0238
0.7939±0.0068
CellFlow
0.7259±0.0203
0.01556±0.00011
0.2563±0.0090
0.8163±0.0182
PerturbDiff
0.6249±0.0871
0.01957±0.00249
0.1717±0.0213
0.7182±0.0273
Sum of singles
0.8333±0.0005
0.01465±0.00001
0.2831±0.0006
0.9292±0.0011
Appendix
Table 9: Perturbation-effect prediction: Norman, held-out double perturbations in K562. All 111 test double perturbations. Values are mean ± SD across five seed means, each averaged over five evaluation splits (Appendix E.4 ).
Model
Pearson Δ↑
MAE ↓
Overlap@N ↑
L1 discrimination ↑
Occam
0.9050±0.0031
0.01139±0.00020
0.3046±0.0074
0.9551±0.0078
TxPert
0.8596±0.0083
0.01425±0.00018
0.3577±0.0019
0.9586±0.0051
STATE
0.7230±0.0362
0.01689±0.00077
0.3029±0.0317
0.8638±0.0196
CellFlow
0.7983±0.0153
0.01453±0.00018
0.2870±0.0090
0.8936±0.0278
PerturbDiff
0.6784±0.0977
0.01933±0.00273
0.1838±0.0247
0.7131±0.0430
Sum of singles
0.8936±0.0004
0.01314±0.00002
0.3053±0.0011
0.9862±0.0003
Appendix
Table 10: Perturbation-effect prediction: Norman, held-out double perturbations in K562. 37 strongly additive double perturbations, without effect-size grouping. Values are mean ± SD across five seed means, each averaged over five evaluation splits (Appendix E.4 ).
Model
Pearson Δ↑
MAE ↓
Overlap@N ↑
L1 discrimination ↑
Occam
0.7869±0.0059
0.01160±0.00020
0.2503±0.0043
0.9047±0.0032
TxPert
0.6907±0.0094
0.01807±0.00030
0.2668±0.0023
0.7903±0.0086
STATE
0.5998±0.0295
0.01715±0.00096
0.2196±0.0171
0.8673±0.0139
CellFlow
0.6724±0.0170
0.01464±0.00051
0.2154±0.0075
0.8795±0.0010
PerturbDiff
0.5714±0.0825
0.01767±0.00248
0.1560±0.0204
0.8466±0.0174
Sum of singles
0.7673±0.0011
0.01499±0.00003
0.2701±0.0006
0.9128±0.0012
Appendix
Table 11: Perturbation-effect prediction: Norman, held-out double perturbations in K562. 37 weakly additive double perturbations, without effect-size grouping. Values are mean ± SD across five seed means, each averaged over five evaluation splits (Appendix E.4 ).
Model
Pearson Δ↑
MAE ↓
Overlap@N ↑
L1 discrimination ↑
Occam
0.8660±0.0036
0.01375±0.00014
0.2869±0.0057
0.8289±0.0280
TxPert
0.7920±0.0067
0.01728±0.00016
0.3459±0.0030
0.7550±0.0111
STATE
0.6760±0.0354
0.01907±0.00082
0.2964±0.0255
0.6507±0.0110
CellFlow
0.7070±0.0293
0.01751±0.00028
0.2663±0.0106
0.6757±0.0263
PerturbDiff
0.6249±0.0833
0.02171±0.00228
0.1753±0.0198
0.5949±0.0291
Sum of singles
0.8391±0.0007
0.01582±0.00002
0.2740±0.0008
0.8886±0.0025
Appendix
Table 12: Perturbation-effect prediction: Norman, held-out double perturbations in K562. 37 intermediate double perturbations, without effect-size grouping. Values are mean ± SD across five seed means, each averaged over five evaluation splits (Appendix E.4 ).
Model
Train FLOPs ↓
Inf. samples/s 1,024 genes ↑
Inf. samples/s 2,048 genes ↑
Inf. samples/s 44k genes ↑
(Peta FLOPs)
B=512
B=1024
B=2048
B=512
B=1024
B=2048
B=32
B=1024
B=2048
scGPT-90M
33,203
1,933
1,955
1,665
–
–
–
–
–
–
STATE-90M
916,752
1,276
1,284
–
566
–
–
–
–
–
TxFM-B
427,050
827
860
–
372
–
–
4
–
–
Occam-90M
327
1,448,625
1,619,255
2,345,992
1,448,625
1,619,255
2,345,992
102,915
1,619,255
2,345,992
Appendix
Table 13: Training compute and inference throughput. All models were trained on the same H100 GPU cluster on DiverseRNA. Training compute is reported as total training operations, in petaFLOPs. Inference throughput is measured on approximately 100,000 profiles on one H100 80GB GPU; entries report profiles per second, so higher is better. Batch size is varied to show throughput scaling, and omitted where the tested configuration does not fit in 80GB memory. The tested scGPT and STATE configurations use maximum sequence lengths of 1,200 and 2,048 gene tokens. The 44k-gene comparison includes Occam and TxFM. These encoder measurements exclude perturbation-effect prediction and pretraining of external gene and protein embeddings.
HepG2
Jurkat
RPE1
( Nadig et al., 2025 )
( Nadig et al., 2025 )
( Replogle et al., 2022 )
Cell line
HepG2
Jurkat
RPE1
Cells
145,473
262,956
247,914
Genes
9,624
8,882
8,749
Perturbations
2,394
2,394
2,394
Control cells
4,976
12,013
11,485
Appendix
Table 14: Perturbation representation benchmark datasets. Each dataset profiles single-gene CRISPR perturbations in a single cell line. Non-targeting guide cells serve as controls.
Pancreas
PBMC-68k
Cells
16,382
68,579
Genes
19,093
32,738
Cell types
14
11
Batches
6
8
Appendix
Table 15: scIB benchmark datasets. Each dataset provides cell-type annotations and batch labels for assessing biological conservation and batch correction.
Model
Perturbation representation ↑
scIB ↑
RPE1
HepG2
Jurkat
Avg.
Panc.
PBMC-68k
Avg.
Random
19.37
19.28
19.15
19.27
43.59
47.71
45.65
scGPT-90M
24.25
24.31
23.47
24.01
37.39
53.66
45.53
scMAE-90M
26.50
25.85
24.77
25.71
47.12
51.70
49.41
STATE-SE-90M
35.86
32.14
30.78
32.93
47.03
48.38
47.71
scVI-110M
35.55
33.29
30.62
33.16
48.72
55.02
51.87
Appendix
Table 16: Representation-learning benchmark aggregates. Trained methods use DiverseRNA ( Kenyon-Dean et al., 2026 ) , with 1.4 million expression profiles; Random is untrained. For each model, we report the perturbation representation score on HepG2, Jurkat, and RPE1, and the scIB score on Pancreas and PBMC-68k. Avg. denotes the arithmetic mean across datasets within each benchmark. Simple baselines are highlighted in blue.
Model
Perturbation representation ↑
scIB ↑
RPE1
HepG2
Jurkat
Avg.
Panc.
PBMC-68k
Avg.
scGPT-43M
25.20
25.01
24.06
24.75
41.89
55.43
48.66
TrFormer-sp
31.50
28.57
27.11
29.06
52.22
49.17
50.70
AIDO.Cell-100M
34.56
32.21
30.90
32.56
44.95
53.23
49.09
STATE-SE-600M
39.54
34.97
33.19
35.90
46.58
50.83
48.71
Appendix
Table 17: Representation-learning benchmark aggregates. These public checkpoints use pretraining corpora larger than DiverseRNA. TrFormer-sp denotes TranscriptFormer-Sapiens. For each model, we report the perturbation representation score on HepG2, Jurkat, and RPE1, and the scIB score on Pancreas and PBMC-68k. Avg. denotes the arithmetic mean across datasets within each benchmark.
Model
LFC- L2↓
DEG-F1 ↑
RPE1
HepG2
Jurkat
Avg.
RPE1
HepG2
Jurkat
Avg.
Pretrained gene representations
Occam-90M
7.37
6.84
6.42
6.87
0.381
0.434
0.385
0.400
STATE-SE-600M
7.40
6.85
6.43
6.89
0.383
0.426
0.388
0.399
AIDO.Cell-100M
7.41
6.87
6.43
6.91
0.366
0.410
0.373
0.383
ESM2
7.42
6.91
6.45
6.93
0.361
0.398
0.369
0.376
Appendix
Table 18: Gene-representation evaluation with phenotype probes. Supervised probes predict knockdown phenotypes from gene embeddings to evaluate representation quality. LFC- L2 is the mean L2 error for predicted log-fold-change vectors (lower is better). DEG-F1 is the per-gene F1 score for differential-expression class prediction (higher is better). Avg. denotes the arithmetic mean across RPE1, HepG2, and Jurkat. Values are point estimates. TrFormer-sp denotes TranscriptFormer-Sapiens. Baselines are highlighted in blue; rows within each group are ordered by average LFC error.
Model
Rel.
Cons.
iLISI
KNN
Lin.
Inv.
Avg.
Random
10.31
0.00
70.34
8.18
2.49
24.88
19.37
scGPT-90M (DiverseRNA)
25.55
3.13
71.13
9.23
4.14
32.31
24.25
scGPT-43M (public)
30.13
3.55
70.14
9.21
5.22
32.93
25.20
scMAE-90M (DiverseRNA)
26.36
2.83
71.00
10.25
10.99
37.57
26.50
TrFormer-sp (public)
36.12
31.30
55.83
11.46
15.72
38.55
31.50
AIDO.Cell-100M (public)
43.21
36.19
69.11
12.31
9.06
37.46
34.56
Appendix
Table 19: Perturbation representation benchmark on RPE1. We report all six task scores used to compute the RPE1 perturbation representation average. Rel. measures recovery of known gene relationships; Cons. measures within-perturbation consistency; iLISI measures batch mixing among cells receiving the same perturbation; KNN and Lin. measure perturbation-label classification with KNN and linear probes; Inv. measures reconstruction of gene expression from the embeddings. Avg. is the arithmetic mean of the six tasks. Higher is better for all metrics. Simple baselines are highlighted in blue; public denotes a public checkpoint.
Model
Rel.
Cons.
iLISI
KNN
Lin.
Inv.
Avg.
Random
9.64
0.00
70.33
5.44
1.13
29.16
19.28
scGPT-90M (DiverseRNA)
29.74
1.59
70.89
6.40
2.37
34.86
24.31
scGPT-43M (public)
31.53
2.38
69.65
6.50
3.35
36.62
25.01
scMAE-90M (DiverseRNA)
26.93
0.38
71.00
8.05
7.36
41.43
25.85
TrFormer-sp (public)
35.24
19.39
53.50
8.48
12.13
42.69
28.57
STATE-SE-90M (DiverseRNA)
40.66
17.51
69.83
9.57
12.95
42.32
32.14
Appendix
Table 20: Perturbation representation benchmark on HepG2. We report all six task scores used to compute the HepG2 perturbation representation average. Rel. measures recovery of known gene relationships; Cons. measures within-perturbation consistency; iLISI measures batch mixing among cells receiving the same perturbation; KNN and Lin. measure perturbation-label classification with KNN and linear probes; Inv. measures reconstruction of gene expression from the embeddings. Avg. is the arithmetic mean of the six tasks. Higher is better for all metrics. Simple baselines are highlighted in blue; public denotes a public checkpoint.
Model
Rel.
Cons.
iLISI
KNN
Lin.
Inv.
Avg.
Random
10.58
0.00
70.63
7.44
1.29
24.98
19.15
scGPT-90M (DiverseRNA)
26.83
0.33
71.12
7.93
3.03
31.55
23.47
scGPT-43M (public)
30.05
0.33
69.93
8.02
3.67
32.35
24.06
scMAE-90M (DiverseRNA)
25.82
3.25
70.00
9.31
4.49
35.79
24.77
TrFormer-sp (public)
32.70
17.85
56.67
8.94
9.57
36.91
27.11
scVI-110M (DiverseRNA)
37.60
15.38
70.22
10.05
9.43
41.06
30.62
Appendix
Table 21: Perturbation representation benchmark on Jurkat. We report all six task scores used to compute the Jurkat perturbation representation average. Rel. measures recovery of known gene relationships; Cons. measures within-perturbation consistency; iLISI measures batch mixing among cells receiving the same perturbation; KNN and Lin. measure perturbation-label classification with KNN and linear probes; Inv. measures reconstruction of gene expression from the embeddings. Avg. is the arithmetic mean of the six tasks. Higher is better for all metrics. Simple baselines are highlighted in blue; public denotes a public checkpoint.
Model
Biological conservation
Batch correction
Final
ASW-L
NMI
ARI
Iso-ASW
Iso-F1
cLISI
Bio
ASW-B
GC
iLISI
PCR
Batch
scIB
scGPT-90M (DiverseRNA)
34.44
25.73
10.08
28.18
10.00
90.31
33.12
47.26
46.54
1.25
80.14
43.80
37.39
scGPT-43M (public)
41.61
35.27
15.28
41.16
5.41
94.80
38.92
65.28
51.19
0.60
68.34
46.35
41.89
Random
48.28
0.49
0.0
48.24
3.48
78.07
29.76
95.39
26.65
35.39
99.93
64.34
43.59
AIDO.Cell-100M (public)
45.35
49.69
22.21
56.97
8.42
88.51
45.19
72.62
86.94
2.15
16.67
44.59
44.95
STATE-SE-600M (public)
51.56
61.83
26.23
51.60
2.47
93.99
47.95
94.33
83.80
0.00
0.00
44.53
46.58
Appendix
Table 22: scIB benchmark on Pancreas. We report all scIB component metrics. Biological conservation metrics are ASW-L, NMI, ARI, Iso-ASW, Iso-F1, and cLISI; their mean is Bio. Batch correction metrics are ASW-B, graph connectivity (GC), iLISI, and PCR; their mean is Batch. The final scIB score is 0.6Bio+0.4Batch . Higher is better for all metrics. Simple baselines are highlighted in blue; public denotes a public checkpoint.
Model
Biological conservation
Batch correction
Final
ASW-L
NMI
ARI
Iso-ASW
Iso-F1
cLISI
Bio
ASW-B
GC
iLISI
PCR
Batch
scIB
Random
49.90
0.16
0.01
49.87
15.43
69.95
30.88
99.10
52.65
52.98
87.06
72.95
47.71
STATE-SE-90M (DiverseRNA)
50.23
45.76
22.09
40.46
35.18
72.80
44.42
97.12
80.46
39.76
0.00
54.33
48.38
TrFormer-sp (public)
49.89
50.11
20.48
50.75
17.61
72.53
43.56
94.72
81.95
53.66
0.00
57.58
49.17
STATE-SE-600M (public)
50.09
45.61
23.29
45.26
41.38
78.60
47.37
99.53
78.19
46.36
0.00
56.02
50.83
scMAE-90M (DiverseRNA)
50.70
5.16
2.51
50.87
24.43
97.35
38.50
98.30
58.65
49.98
79.06
71.50
51.70
Appendix
Table 23: scIB benchmark on PBMC-68k. We report all scIB component metrics. Biological conservation metrics are ASW-L, NMI, ARI, Iso-ASW, Iso-F1, and cLISI; their mean is Bio. Batch correction metrics are ASW-B, graph connectivity (GC), iLISI, and PCR; their mean is Batch. The final scIB score is 0.6Bio+0.4Batch . Higher is better for all metrics. Simple baselines are highlighted in blue; public denotes a public checkpoint.
Australian Institute for Machine Learning, Adelaide University, Australia · Responsible AI Research Centre, Australia · College of Computer Science and Artificial Intelligence, Fudan University, China +3
School of Computer Science and Engineering, South China University of Technology, Guangzhou, China · School of Future Technology, South China University of Technology, Guangzhou, China