fluorometer turner designs model-10 Search Results


bd-10  (Becton Dickinson)
90
Becton Dickinson bd-10
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Bd 10, supplied by Becton Dickinson, used in various techniques. Bioz Stars score: 90/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/bio_rxiv__2021__02__10__430606-137-12-12?v=Becton+Dickinson
Average 90 stars, based on 1 article reviews
bd-10 - by Bioz Stars, 2026-08
90/100 stars
  Buy from Supplier
core diabetes model  (IQVIA Inc)
90
IQVIA Inc core diabetes model
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Core Diabetes Model, supplied by IQVIA Inc, used in various techniques. Bioz Stars score: 90/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/pm32887705-35-30-29?v=IQVIA+Inc
Average 90 stars, based on 1 article reviews
core diabetes model - by Bioz Stars, 2026-08
90/100 stars
  Buy from Supplier
sugen hypoxia rat model  (Sugen Inc)
86
Sugen Inc sugen hypoxia rat model
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Sugen Hypoxia Rat Model, supplied by Sugen Inc, used in various techniques. Bioz Stars score: 86/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/10__1164_slash_rccm__202102___0446oc-244-15-15?v=Sugen+Inc
Average 86 stars, based on 1 article reviews
sugen hypoxia rat model - by Bioz Stars, 2026-08
86/100 stars
  Buy from Supplier
micro-orifice uniform deposition impactor moudi model 10  (MSP Corp)
90
MSP Corp micro-orifice uniform deposition impactor moudi model 10
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Micro Orifice Uniform Deposition Impactor Moudi Model 10, supplied by MSP Corp, used in various techniques. Bioz Stars score: 90/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/pmc04736724-92-1-8?v=MSP+Corp
Average 90 stars, based on 1 article reviews
micro-orifice uniform deposition impactor moudi model 10 - by Bioz Stars, 2026-08
90/100 stars
  Buy from Supplier
anova model 10 water circulator  (Anova Scientific)
90
Anova Scientific anova model 10 water circulator
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Anova Model 10 Water Circulator, supplied by Anova Scientific, used in various techniques. Bioz Stars score: 90/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/pm37872294-190-15-13?v=Anova+Scientific
Average 90 stars, based on 1 article reviews
anova model 10 water circulator - by Bioz Stars, 2026-08
90/100 stars
  Buy from Supplier
clark- type electrode digital model 10  (Rank Brothers Ltd)
90
Rank Brothers Ltd clark- type electrode digital model 10
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Clark Type Electrode Digital Model 10, supplied by Rank Brothers Ltd, used in various techniques. Bioz Stars score: 90/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/pm22982531-50-7-13?v=Rank+Brothers+Ltd
Average 90 stars, based on 1 article reviews
clark- type electrode digital model 10 - by Bioz Stars, 2026-08
90/100 stars
  Buy from Supplier
ph meter corning 10  (Corning Life Sciences)
90
Corning Life Sciences ph meter corning 10
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Ph Meter Corning 10, supplied by Corning Life Sciences, used in various techniques. Bioz Stars score: 90/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/10__59493_slash_ajopred_slash_2024__1__7-45-31-31?v=Corning+Life+Sciences
Average 90 stars, based on 1 article reviews
ph meter corning 10 - by Bioz Stars, 2026-08
90/100 stars
  Buy from Supplier
catheter  (ADInstruments)
94
ADInstruments catheter
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Catheter, supplied by ADInstruments, used in various techniques. Bioz Stars score: 94/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/pmc08037792-172-9-18?v=ADInstruments
Average 94 stars, based on 1 article reviews
catheter - by Bioz Stars, 2026-08
94/100 stars
  Buy from Supplier
gamma counter icn biomedicals model 10/600 plus  (ICN Biomedicals)
90
ICN Biomedicals gamma counter icn biomedicals model 10/600 plus
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Gamma Counter Icn Biomedicals Model 10/600 Plus, supplied by ICN Biomedicals, used in various techniques. Bioz Stars score: 90/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/10__1074_slash_jbc__272__4__2346-73-8-10?v=ICN+Biomedicals
Average 90 stars, based on 1 article reviews
gamma counter icn biomedicals model 10/600 plus - by Bioz Stars, 2026-08
90/100 stars
  Buy from Supplier
headplate headplate model 10  (Neurotar Ltd)
90
Neurotar Ltd headplate headplate model 10
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Headplate Headplate Model 10, supplied by Neurotar Ltd, used in various techniques. Bioz Stars score: 90/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/pmc12202027-395-27-32?v=Neurotar+Ltd
Average 90 stars, based on 1 article reviews
headplate headplate model 10 - by Bioz Stars, 2026-08
90/100 stars
  Buy from Supplier
cgmd simulations kremer–grest model  (GRest Inc)
90
GRest Inc cgmd simulations kremer–grest model
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Cgmd Simulations Kremer–Grest Model, supplied by GRest Inc, used in various techniques. Bioz Stars score: 90/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/10__1016_slash_j__polymer__2022__124637-26-6-4?v=GRest+Inc
Average 90 stars, based on 1 article reviews
cgmd simulations kremer–grest model - by Bioz Stars, 2026-08
90/100 stars
  Buy from Supplier
video camera sony, model 10.1 megapixels, optical zoom 3x  (Sony)
90
Sony video camera sony, model 10.1 megapixels, optical zoom 3x
Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and <t>BD-10</t> model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.
Video Camera Sony, Model 10.1 Megapixels, Optical Zoom 3x, supplied by Sony, used in various techniques. Bioz Stars score: 90/100, based on 1 PubMed citations. ZERO BIAS - scores, article reviews, protocol conditions and more
https://www.bioz.com/product/fluorometer+turner+designs+model-10/10__1016_slash_j__lwt__2015__03__059-78-8-10?v=Sony
Average 90 stars, based on 1 article reviews
video camera sony, model 10.1 megapixels, optical zoom 3x - by Bioz Stars, 2026-08
90/100 stars
  Buy from Supplier

Image Search Results


Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and BD-10 model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.

Journal: bioRxiv

Article Title: NeuronMotif: Deciphering transcriptional cis-regulatory codes from deep neural networks

doi: 10.1101/2021.02.10.430606

Figure Lengend Snippet: Motif discovery performance for different layers in different models. a , Accuracy analysis for discovered motifs of different models. Three columns of box plots describe similarity between neuron motifs and JASPAR motifs in Basset, BD-5 and BD-10 model respectively. For each selected layer in the model, -log10(q-values) distribution of the top 100 JASPAR motifs matched neuron (q-value < 0.1) are shown with box and jittering points. The color of the box means the applied interpretation method. In the first row, it shows the result of input (first) layer of Basset model and shallow layers in BD-5 and BD-10 model with similar receptive field sizes (18 bp) of the Basset input layer neuron (19 bp). In the second row, it shows the convolutional output layer, the convolutional layer in the front of dense layer, result of the three models. b , The number of motifs discovered (q-value < 0.001) from the neuron in convolutional output layer of Basset, BD-5 and BD-10 model. c , The number of motif discovered (q-value < 0.01) from the neuron in layer 3 of Basset model using different interpretation methods including Kelley et al., Alipanahi et al. and NeuronMotif. d , Discovered motifs from the neuron of top convolutional layer in BD-10 model (q-value < 0.01). These motifs can be matched to JASPAR database. Only the one with smallest q-value for each JASPAR motif is shown.

Article Snippet: For example, we built a motif dictionary from layer 10 (L10) in BD-10 model.

Techniques:

NeuronMotif diagnose model defects and guide DCNN architecture design for better performance. a , Dead kernel definition. Dead neuron (pink) activation value distribution is negative. It will be filtered by ReLU activation function. The output of dead neuron is zero. The downstream neuron output does not depend on this neuron. b , c , Diagnosis of motif mixture in the decoupled motifs from Basset, BD-10, DeepSEA and DD10 model. Each point is a decoupled motif generate by a sample set of sequence. The points of motifs generated by the sample set with less than 100 sequences are marked by red color. Otherwise, they are marked by blue color. The distribution of the max activation value is used to show if the relative max activation values of most of the motifs are too low. b , Diagnosis of models trained by DeepSEA dataset c , Diagnosis of models trained by Basset dataset. Only the max activation value of the decoupled motifs in are significantly higher than the decoupled motifs of other neurons in layer 3 of Basset-3 model. d , The meaning of each region in the sub-plots of b , c . e , Schematic for receptive field coupling of previous layer neuron in the neuron sub-structure. f,g , Use AUPRC as an indicator to compare the prediction performance of models. For each model pair, one-sided t-test of Δ AUPRC = AUPRC y-axis – AUPRC x-axis is used to access model performance difference level. f , Comparison between DeepSEA and DD-10 models. g , Comparison between Basset and BD-10 models.

Journal: bioRxiv

Article Title: NeuronMotif: Deciphering transcriptional cis-regulatory codes from deep neural networks

doi: 10.1101/2021.02.10.430606

Figure Lengend Snippet: NeuronMotif diagnose model defects and guide DCNN architecture design for better performance. a , Dead kernel definition. Dead neuron (pink) activation value distribution is negative. It will be filtered by ReLU activation function. The output of dead neuron is zero. The downstream neuron output does not depend on this neuron. b , c , Diagnosis of motif mixture in the decoupled motifs from Basset, BD-10, DeepSEA and DD10 model. Each point is a decoupled motif generate by a sample set of sequence. The points of motifs generated by the sample set with less than 100 sequences are marked by red color. Otherwise, they are marked by blue color. The distribution of the max activation value is used to show if the relative max activation values of most of the motifs are too low. b , Diagnosis of models trained by DeepSEA dataset c , Diagnosis of models trained by Basset dataset. Only the max activation value of the decoupled motifs in are significantly higher than the decoupled motifs of other neurons in layer 3 of Basset-3 model. d , The meaning of each region in the sub-plots of b , c . e , Schematic for receptive field coupling of previous layer neuron in the neuron sub-structure. f,g , Use AUPRC as an indicator to compare the prediction performance of models. For each model pair, one-sided t-test of Δ AUPRC = AUPRC y-axis – AUPRC x-axis is used to access model performance difference level. f , Comparison between DeepSEA and DD-10 models. g , Comparison between Basset and BD-10 models.

Article Snippet: For example, we built a motif dictionary from layer 10 (L10) in BD-10 model.

Techniques: Activation Assay, Sequencing, Generated