Table 1.
Current machine learning techniques, software platforms, and data resources used for automatic neuronal classification.
| Refs | ML-technique | Software platform | Sample size | Data domains | Availability |
|---|---|---|---|---|---|
| [5] | ANOVA, hierarchical clustering, t-test, Gene Ontology |
MAS, dChip, R, Python, IgorPro |
36 microarrays of mouse forebrain cells (12 types) |
13,232 mRNA transcripts | 1-C* |
| [6] | Pearson correlation, ANOVA, t-test | SPSS, Prism | 520 crab stomatogastric ganglion cells (6 types) |
6 genes qPCR expression | 2-C |
| [7] | Intensity levels, SAM method, hierarchical clustering (complete linkage) |
MAS, dChip, SAM, Bioconductor, IPA, David |
29 mice forebrain astrocytes, neurons and oligodendrocytes (9 subtypes) |
45,037 mRNA probe sets | 1-C |
| [8] | Memory-guided saccade task, visuomovement index for each neuron, coefficient of variation, Wilcoxon ranksum test, Bonferroni correction |
R | 94 frontal eye field cells (3 subtypes of cells, 4 male macaques) |
1 electrophysiological (width of spikes) |
2-B |
| [9] | Hierarchical clustering (Ward’s linkage), k- means, Mann–Whitney U test, silhouette analysis |
Statistica, Matlab | 68 out of 200 cortical interneurons layers I to IV Wistar rat (6 subtypes) |
1 positional, 12 morphological, 32 electrophysiological, 10 molecular features |
2-B-C |
| [10] | Hierarchical clustering (Spearman correlation distance), t-test, fold-change, Gene Ontology |
MAS, GenMAPP, IPA |
16 microarrays of P8 mice tissues (8 subplate, 8 Layer VI cells) |
39,000 mRNA transcripts | 3-C |
| [11] | PCA, hierarchical clustering (Ward’s linkage), k-means, Mann Whitney U test, silhouette analysis |
Statistica, Clustan | 59 SOM and cortical mouse interneurons |
19 electrophysiological, 67 morphological features |
1-C |
| [12] | Hierarchical clustering, naïve Bayes, C4.5, kNN, MLP, Logistic regression, PCA, FSS |
R, WEKA | 327 mouse neocortex cells (128 pyramidal and 199 interneurons) |
65 morphological features | 1-C |
| [13] | Seven clustering algorithms: reported results from four of them (Fuzzy Gustafson-Kessel, k-means, PAM, affinity) |
Matlab | 471 mouse retinal ganglion cells (12 subtypes) |
5 electrophysiological | 3-D |
| [14] | Hierarchical clustering, fold-change, ANOVA, Gene Ontology |
Affymetrix Power Tools |
195 mouse microarrays (64 cell types) |
22,690 mRNA probe sets | 1-B |
| [15] | Coefficient of variation of interspike interval, hierarchical clustering, correlation, cross-covariance, t-tests |
Matlab, MClust | 207 ventral tegmental cells, male Long-Evans rats |
Action potential firing, dopamine receptor pharmacology |
2-B |
| [16] | Mutual information, kNN, area under the receiver operating characteristic curve (AUC) |
Matlab | 60 Drosophila melanogaster neuroblast clones |
350 whole brain images neuroblast clones |
1-A* |
| [17] | Expert crowdsourcing, k-means, Bayesian networks |
Matlab, WEKA | 320 cortical interneurons, mouse, rat, rabbit, cat, monkey and human |
6 polled axonal-derived morphological features |
1-D* |
| [18] | Unsupervised fuzzy sets clustering | Matlab | 200 cortical interneurons Wistar rat (6 subtypes) |
1 positional, 32 electrophysiological, 10 molecular features |
2-B |
| [19] | PCA, LDA, FSS, Gaussian distributions, k- means, Rand index |
Matlab | 466 P12-16 Wistar rat cortical interneurons (8 subtypes based on PING criteria [4]) |
38 electrophysiological | 3-D |
| [20] | Idem as in [9] | Statistica, Matlab | 123 interneurons of Wistar rat barrel cortex (see [9]) |
1 positional, 32 electrophysiological, 11 molecular features |
3-B-C |
| [21] | Affinity propagation clustering | Neurolucida, Matlab | 337 P13-25 mouse interneurons (4 subtypes: PV+ basket, PV+ chandelier, SOM+ Martinotti, SOM+ non- Martinotti) |
67 morphological, 20 electrophysiological features |
1-B |
| [22] | Bayesian networks, k-means, Bayesian multinets |
GeNIe, R | 320 cortical interneurons, mouse, rat, rabbit, cat, monkey and human |
Idem as in [17] | 1-C* |
| [23] | Harmonic co-clustering, wavelet smoothing | FARSIGHT Trace Editor |
728 mouse necortex pyramidal (6 subtypes), 502 axonal rat hippocampus (4 subtypes) |
130 morphological features | 1-A* |
| [24] | Hierarchical clustering | Matlab | 363 mouse retinal ganglion (15 subtypes) |
48,000 morphological voxels | 1-A |
| [25] | EM clustering, manual classification | KlustaKwik, SpikeDetekt, NeuroSuite |
6,732 Long-Evans rat hippocampus and EC cells (12 subtypes) |
31-127 electrophysiological channels |
1-A* |
| [26] | Hierarchical clustering, kNN | R, NBLAST | 16,129 Drosophila melanogaster neurons (1,052 subtypes) |
~ 1,070 spatial points | 1-A* |
| [39] | Bayesian network classifiers, FSS | R, WEKA | 237 cortical interneurons, mouse, rat, rabbit, cat, monkey and human |
214 morphological features | 1-A |
| [48] | Affinity propagation clustering, kNN | Matlab, NeuTu | 379 Drosophila melanogaster optic medulla (89 subtypes) |
10xN (user fixed) branch density similarities from morphological skeletons |
1-A |
| [59] | Hierarchical clustering (Ward’s linkage), ANOVA, Fisher test |
Statistica | 77 dorsolateral prefrontal cortex pyramidal cells, male long-tailed macaque (4 subtypes) |
16 electrophysiological parameters (membrane properties) |
3-C |
| [60] | Hierarchical clustering (Ward’s linkage), k- means, PCA, t-test, silhouette analysis, bootstrap test |
IGOR Pro, R | 82 mouse interneurons (4 subtypes) |
17 electrophysiological parameters (membrane properties) |
3-A-C |
| [84] | Hierarchical clustering (Ward’s linkage), Principal Factor Analysis, Spearman correlation, Kruskal-Wallis test, Mann- Whitney U test |
Stimfit, Mathematica, Matlab, SPSS |
114 P18-25 mouse dentate gyrus GABAergic interneurons (5 subtypes) |
31 morphological, 34 electrophysiological features |
1-C |
1 – Data available online
2 – Data available upon request
3 – Data not available
A – Software code available online
B – Software code available upon request
C – Commercial software without coding
D – Software or specific code not available
*
– Data and/or software dedicated website