Journal: eLife

Article Title: Noisy neuronal populations effectively encode sound localization in the dorsal inferior colliculus of awake mice

doi: 10.7554/eLife.97598

Figure Lengend Snippet: ( A ) Schematic representation of the decoding strategy using multi-class classification on the recorded simultaneous population responses. ( B ) Cumulative distribution plots of the absolute cross-validated single-trial prediction errors obtained using naive Bayes classification (blue) and chance level distribution associated with our stimulation paradigm obtained by considering all possible prediction errors for the 13 azimuths tested (gray). K.S.: Kolmogorov-Smirnov test, n.s.: (p>0.05). ( C ) Schematic representation of the decoding strategy using the first PCs of the recorded population responses. Inset: % of explained variance obtained using PCA for dimensionality reduction on the complete population responses. Median (blue line) and median absolute deviation (shaded blue area) are plotted for (n=12 mice/imaging sessions). ( D ) Same as ( B ) but for decoding different numbers of first PCs from the recorded complete population responses. ( E ) Significance of classification performance with respect to chance level for different numbers of first PCs, determined via Kolmogorov-Smirnov tests with Sidak correction for multiple comparisons. Arrowhead indicates model loss of performance associated with fitting more parameters for a larger feature space (# PCs) with the same dataset size (# trials collected).

Article Snippet: IC population response classification (decoding) was performed using the naive Bayes classifier algorithm implemented in MATLAB, using the fitcnb function from the statistics and machine learning toolbox.

Techniques: Imaging

Journal: eLife

Article Title: Noisy neuronal populations effectively encode sound localization in the dorsal inferior colliculus of awake mice

doi: 10.7554/eLife.97598

Figure Lengend Snippet: ( A ) Histogram of the nS/N ratios from the recorded units across mice during sound stimulation or during the inter trial periods without sound stimulation (on going). ( B ) Representative stimulus azimuth tuning curves from units with significant median response tuning detected using non-parametric one-way ANOVA (Kruskal-Wallis test). Median and absolute median deviation are plotted. The imaging depth from the corresponding units is displayed in gray. Azimuth selectivity is color-coded based on . ( C ) Percentage of the simultaneously recorded units across mice that showed significant median response tuning, compared to false positive detection rate ( α =0.05, chance level). ( D ) Response dependency to stimulus azimuth, determined via χ 2 tests (see Materials and methods), for simultaneously recorded units ranked in descending order of significance. Left inset: Representative responses from the top ranked 7 units with significant response dependency to stimulus azimuth. Response amplitudes are displayed with a continuous trace for visualization purposes, the displayed response order was sorted as a function of stimulus azimuth and does not represent the experimental stimulus delivery order (random). Right inset: Same as ( A ) but for the subset of units displaying response dependency to stimulus azimuth. ( E ) Percentage of the simultaneously recorded units across mice that showed significant response dependency to stimulus azimuth, compared to false positive detection rate ( α =0.05, chance level). ( F ) Schematic representation of the decoding strategy using the top ranked units from the recorded population responses. ( G ) Top: Cumulative distribution plot of the absolute cross-validated single-trial prediction errors obtained with a Bayes classifier (N. Bayes, naive approximation for computation efficiency). The number of top ranked units considered for decoding their simultaneously recorded single-trial population response patterns is color coded from cyan (4 top ranked units) to purple (10 top ranked units) and the chance level distribution associated to our stimulation paradigm, obtained by considering all possible prediction errors for the 13 azimuths tested, is displayed in gray. Bottom: Significance of classification performance with respect to chance level for 4–30 decoded top ranked units, determined via Kolmogorov-Smirnov tests with Sidak correction for multiple comparisons. Arrowhead indicates model loss of performance associated with fitting more parameters for a larger feature space (# units) with the same dataset size (# trials collected).

Article Snippet: IC population response classification (decoding) was performed using the naive Bayes classifier algorithm implemented in MATLAB, using the fitcnb function from the statistics and machine learning toolbox.

Techniques: Imaging

Journal: eLife

Article Title: Noisy neuronal populations effectively encode sound localization in the dorsal inferior colliculus of awake mice

doi: 10.7554/eLife.97598

Figure Lengend Snippet: ( A ) Simplified schematic representation of the possible effects from (positive) noise correlations on the response separability of a theoretical population consisting of 2 units, and within class randomization strategy to model decorrelated datasets lacking noise correlations. ( B, C ) Left top: Representative correlation matrices of pairwise correlations between the responses of top ranked units detected in simultaneous recordings during sound stimuli for representative azimuths. The simultaneously imaged units are sorted in the correlation matrices based on cross validated hierarchical clustering (see Materials and methods). Left bottom: Distribution histograms for the pairwise correlation coefficients (Kendall tau) from pairs of simultaneously recorded top ranked units across mice (blue) compared to the chance level distribution obtained through randomization of the temporal structure of each unit’s activity to break correlations (purple). Vertical dashed lines show the medians of these distributions. * : p<0.05, *** : p<0.0001, Kolmogorov-Smirnov with Sidak. Right: Cumulative distribution plots of the absolute cross-validated single-trial prediction errors obtained using a Bayes classifier (naive approximation for computation efficiency) to decode the single-trial response patterns from the 6 (neuropixels) or 7 (sTeFo 2 P imaging) top ranked units in the simultaneously acquired datasets across mice (cyan), modeled decorrelated datasets (orange) and the chance level distribution associated with our stimulation paradigm (gray).

Article Snippet: IC population response classification (decoding) was performed using the naive Bayes classifier algorithm implemented in MATLAB, using the fitcnb function from the statistics and machine learning toolbox.

Techniques: Activity Assay, Imaging

Journal: Cell systems

Article Title: Prediction of compound synergism from chemical-genetic interactions by machine learning

doi: 10.1016/j.cels.2015.12.003

Figure Lengend Snippet: A Chemical Genetic Matrix (CGM). (A) Data generation and analysis workflow. 4915 unique molecules from 4 different chemical libraries were screened against a panel of 195 S. cerevisiae deletion strains (termed sentinels) to identify compounds that inhibit the growth of specific deletion strains (termed cryptagens). Pairwise combinations of 128 structurally diverse cryptagens from the CGM were screened in a 128×128 cryptagen matrix (CM) to identify synergistic compound pairs. The CGM dataset and chemical structural features were used to build a Naïve Bayes multi-class learner (NBL) to predict compound activity likelihoods for each sentinel strain. A graph-based algorithm was used to integrate chemical-genetic and genetic interactions to predict compound targets, based on either CGM interaction data (SONARG) or NBL likelihood scores (SONARGN). A random forest-based machine learning algorithm was used to enhance synergy prediction using the CM as training data, based on NBL likelihoods with (SONARGNR) or without (SONARNR) genetic interaction constraints. Predicted synergistic combinations were tested in S. cerevisiae, fungal pathogens and human cell lines. (B) GO SLIM categories represented by sentinel deletion strains. Color indicates significance of gene enrichment based on hypergeometric test. Numbers indicate genes in each category. (C) Heatmap of chemical-genetic interactions in the CGM. Compound activities versus sentinel strains are shown for each individual library screened in this study. (D) Compound activities (Zscore < −4) across sentinel strains. 1221 cryptagens that sensitized > 4 and < 2/3 of all deletion strains are indicated. Inset: Median growth inhibition across all sentinel screens for each compound. See also Figures S1, S2.

Article Snippet: Structural characteristics of each cryptagen were represented by Extended-Connectivity Fingerprints (ECFP4, see Methods) and combined with CGM data using a Naïve Bayes multi-class algorithm (SciTegic Pipeline Pilot, see Methods and Fig. S5A ) to predict compound activities towards each sentinel strain.

Techniques: Activity Assay, Inhibition

Journal: Cell systems

Article Title: Prediction of compound synergism from chemical-genetic interactions by machine learning

doi: 10.1016/j.cels.2015.12.003

Figure Lengend Snippet: Synergy Prediction Based on Chemical-Genetic and Genetic Interactions. (A) Deletion strains are sensitized to specific cryptagens. (B) Underlying genetic interaction network. (C) SONARG integrates chemical-genetic and genetic interactions to predict chemical synergies. Sentinel strains sensitive to cryptagen c represent first order connections s. Second order connections t are inferred from genetic interactions of sentinel strains and ranked by interactions with sentinel strains in s. Edge weights between target spaces ti and tj are based on genetic interaction counts. See Methods for details. (D) PCA biplot of loadings for 7 SONARG parameters in comparison to Bliss independence values from the CM. Abbreviations: sgi, shared genetic interactions between deletion strains for each compound pair; pval, p-value; hs, high sum on V vertices for x and y and E edges between x and y. (E) Naïve Bayes multi-class likelihoods from the CGM. ECFP4 fingerprints for all compounds and activity probabilities for each feature are calculated for all sentinel strains. The integrated probability for compound activity across all features and classes is represented as a likelihood score. (F) Heatmap of CGM based on NBL likelihoods. (G) PCA biplot for SONARGN parameters. (H) Receiver-operator characteristics (ROC) for the single property Exy (AUC = 0.64) and for synergy scores based on SONARGNR parameters (AUC = 0.87). Inset: Precision-recall plot for SONARGNR model. (I) Distribution of SONARGNR scores for synergistic and non-synergistic pairs based on CM data. See also Figures S3–S6.

Article Snippet: Structural characteristics of each cryptagen were represented by Extended-Connectivity Fingerprints (ECFP4, see Methods) and combined with CGM data using a Naïve Bayes multi-class algorithm (SciTegic Pipeline Pilot, see Methods and Fig. S5A ) to predict compound activities towards each sentinel strain.

Techniques: Comparison, Activity Assay

Journal: Cell systems

Article Title: Prediction of compound synergism from chemical-genetic interactions by machine learning

doi: 10.1016/j.cels.2015.12.003

Figure Lengend Snippet: Random Forest-Based Learner for Synergy Prediction Based on Chemical-Genetic Interactions and Chemical Structural Features. (A) ROC for synergy prediction with SONARNR model. Inset: Precision-recall plot. (B) Scatterplot of Bliss independence values and SONARNR synergy scores. (C) Naïve Bayes classes of top-ranked deletion strains that predict synergistic interactions. Mean decrease in Gini represents the influence of variables in partitioning the data into defined classes. (D) Sentinel strains associated with synergistic interactions predicted by SONARNR. Genes are grouped by biological processes. Edge weights are determined by NBL likelihood of two genes being among the top three sensitive genes for synergistic pairs, corrected by subtraction of weights for the same graph generated from 730 non-synergistic pairs. (E) Corresponding edge weights for genetic interactions between strains for graph in panel D. See also Figures S8, S9.

Article Snippet: Structural characteristics of each cryptagen were represented by Extended-Connectivity Fingerprints (ECFP4, see Methods) and combined with CGM data using a Naïve Bayes multi-class algorithm (SciTegic Pipeline Pilot, see Methods and Fig. S5A ) to predict compound activities towards each sentinel strain.

Techniques: Generated