weng-lab · thomas-reimonn · Feb 5, 2023 · Feb 5, 2023 · Feb 5, 2023 · Feb 6, 2023
diff --git a/.github/workflows/latest-docker-image.yml → ...s/chrombpnet-with-bigwig-docker-image.yml b/.github/workflows/latest-docker-image.yml → ...s/chrombpnet-with-bigwig-docker-image.yml
@@ -9,7 +9,7 @@ jobs:
   build:
 
     runs-on: ubuntu-latest
-    if: GitHub.ref == 'refs/heads/master'
+    if: GitHub.ref == 'refs/heads/chrombpnet_with_bigwig'
     steps:
     - uses: actions/checkout@v2
     - name: Build the Docker image
@@ -21,8 +21,8 @@ jobs:
 
     - name: Build the Docker image
 
-      run: docker build . --file dockerfile --tag kundajelab/chrombpnet:latest
+      run: docker build . --file dockerfile --tag kundajelab/chrombpnet:chrombpnet_with_bigwig
 
     - name: Docker Push
-      run: docker push kundajelab/chrombpnet:latest
+      run: docker push kundajelab/chrombpnet:chrombpnet_with_bigwig
 
diff --git a/chrombpnet/CHROMBPNET.py b/chrombpnet/CHROMBPNET.py
@@ -90,7 +90,7 @@ def main():
 
 		modisco_command = "modisco motifs -i {} -n 50000 -o {} -w 500".format(args.h5py,args.output_prefix+"_modisco.h5")
 		os.system(modisco_command)
-		modisco_command = "modisco report -i {} -o {} -m {}".format(args.output_prefix+"_modisco.h5",args.output_prefix+"_reports",meme_dir)
+		modisco_command = "modisco report -i {} -o {} -m {}".format(args.output_prefix+"_modisco.h5",args.output_prefix+"_reports",meme_file)
 		os.system(modisco_command)
 
 		import chrombpnet.evaluation.modisco.convert_html_to_pdf as convert_html_to_pdf

diff --git a/chrombpnet/helpers/generate_reports/make_html.py b/chrombpnet/helpers/generate_reports/make_html.py
@@ -168,6 +168,16 @@ def qc_report(fpx,prefix,data_type):
 	#table_profile = open(os.path.join(prefix,"evaluation/modisco_profile/motifs.html")).read().replace("./","./modisco_profile/").replace("width=\"240\"","class=\"cover\"").replace("border=\"1\" class=\"dataframe\"","").replace(">pos_patterns.pattern",">pos_").replace(">neg_patterns.pattern",">neg_").replace("modisco_cwm_fwd","cwm_fwd").replace("modisco_cwm_rev","cwm_rev").replace("num_seqlets","NumSeqs").replace("dataframe","new")
 	#table_counts = open(os.path.join(prefix,"auxiliary/interpret_subsample/modisco_counts/motifs.html")).read().replace("./","./modisco_counts/").replace("width=\"240\"","class=\"cover\"").replace("border=\"1\" class=\"dataframe\"","").replace(">pos_patterns.pattern",">pos_").replace(">neg_patterns.pattern",">neg_").replace("modisco_cwm_fwd","cwm_fwd").replace("modisco_cwm_rev","cwm_rev").replace("num_seqlets","NumSeqs")
 
+	html_perf = f'''
+
+			<body style="font-size:20px;">
+				<h3>{chrombpnet_model_perf_hed}</h3>
+				<p>{chrombpnet_model_perf_text}</p>
+				{pdf.to_html(classes='mystyle')}
+				{pdf1.to_html(classes='mystyle')}
+			</body>	
+		'''
+
 	if data_type == "ATAC":
 		tn5_1 = os.path.join("./","{}chrombpnet_nobias.tn5_1.footprint.png".format(fpx))
 		tn5_2 = os.path.join("./","{}chrombpnet_nobias.tn5_2.footprint.png".format(fpx))
@@ -202,6 +212,18 @@ def qc_report(fpx,prefix,data_type):
 						</table>
 					</body>
 		'''	
+		html_motifs = f'''			
+				<body style="font-size:20px;">
+					<h3>{tf_hed}</h3>
+					<p>{tf_text_profile}</p>
+				</body>
+				<body>
+					 {table_profile}
+				</body>
+			'''
+		html = html_perf+html_table+html_motifs
+		return html.format(tn5_1=tn5_1,tn5_2=tn5_2,tn5_3=tn5_3,tn5_4=tn5_4,tn5_5=tn5_5)
+
 	elif data_type == "DNASE":
 		dnase_1 = os.path.join("./","{}chrombpnet_nobias.dnase_1.footprint.png".format(fpx))
 		dnase_2 = os.path.join("./","{}chrombpnet_nobias.dnase_2.footprint.png".format(fpx))
@@ -227,18 +249,22 @@ def qc_report(fpx,prefix,data_type):
 						</table>
 					</body>	
 		'''
+		html_motifs = f'''			
+				<body style="font-size:20px;">
+					<h3>{tf_hed}</h3>
+					<p>{tf_text_profile}</p>
+				</body>
+				<body>
+					 {table_profile}
+				</body>
+			'''
+		html = html_perf+html_table+html_motifs
+		return html.format(dnase_1=dnase_1,dnase_2=dnase_2)
+
 	else:
 		print("Unknown data type: "+data_type)
 
-	html_perf = f'''
-
-			<body style="font-size:20px;">
-				<h3>{chrombpnet_model_perf_hed}</h3>
-				<p>{chrombpnet_model_perf_text}</p>
-				{pdf.to_html(classes='mystyle')}
-				{pdf1.to_html(classes='mystyle')}
-			</body>	
-		'''
+
 
 # 	html_motifs = f'''			
 # 			<body style="font-size:20px;">
@@ -255,17 +281,6 @@ def qc_report(fpx,prefix,data_type):
 # 				 {table_counts}
 # 			</body>
 # 		'''					
-	html_motifs = f'''			
-			<body style="font-size:20px;">
-				<h3>{tf_hed}</h3>
-				<p>{tf_text_profile}</p>
-			</body>
-			<body>
-				 {table_profile}
-			</body>
-		'''
-	html = html_perf+html_table+html_motifs
-	return html.format(tn5_1=tn5_1,tn5_2=tn5_2,tn5_3=tn5_3,tn5_4=tn5_4,tn5_5=tn5_5)
 
 def main(args):
 

diff --git a/chrombpnet/helpers/hyperparameters/find_bias_hyperparams.py b/chrombpnet/helpers/hyperparameters/find_bias_hyperparams.py
@@ -104,6 +104,10 @@ def main(args):
     print("counts_loss_weight:", counts_loss_weight)
     assert(counts_loss_weight != 0)
 
+    if counts_loss_weight < 1.0:
+        counts_loss_weight = 1.0
+        print("WARNING: you are training on low-read depth data")
+
     # store the parameters being used  - in a TSV file
     file = open("{}bias_data_params.tsv".format(args.output_prefix),"w")
     file.write("\t".join(["counts_sum_min_thresh", str(round(lower_thresh,2))]))

diff --git a/chrombpnet/parsers.py b/chrombpnet/parsers.py
@@ -46,10 +46,13 @@ def general_training_args(required_train, optional_train):
 
         	required_train.add_argument('-g','--genome', required=True, type=str, help="reference genome fasta file")
        		required_train.add_argument('-c', '--chrom-sizes', type=str, required=True, help="Chrom sizes file")
+
         	group = required_train.add_mutually_exclusive_group(required=True)
+
        		group.add_argument('-ibam', '--input-bam-file', type=str, help="Input BAM file")
         	group.add_argument('-ifrag', '--input-fragment-file', type=str, help="Input fragment file")
         	group.add_argument('-itag', '--input-tagalign-file', type=str, help="Input tagAlign file")
+        	group.add_argument("-ibw", "--bigwig", type=str, help="Input bigwig file of observed data")
         	required_train.add_argument('-o', '--output-dir', type=str, required=True, help="Output dir (path/to/output/dir)")
         	required_train.add_argument('-d', '--data-type', required=True, type=str, choices=['ATAC', 'DNASE'], help="assay type")
         	required_train.add_argument("-p", "--peaks", type=str, required=True, help="10 column bed file of peaks. Sequences and labels will be extracted centered at start (2nd col) + summit (10th col).")

diff --git a/chrombpnet/pipelines.py b/chrombpnet/pipelines.py
@@ -12,17 +12,21 @@ def chrombpnet_train_pipeline(args):
 		fpx = args.file_prefix+"_"
 	else:
 		fpx = ""
-
-	# Shift bam and convert to bigwig
-	import chrombpnet.helpers.preprocessing.reads_to_bigwig as reads_to_bigwig	
-	args.output_prefix = os.path.join(args.output_dir,"auxiliary/{}data".format(fpx))
-	args.plus_shift = None
-	args.minus_shift = None
-	reads_to_bigwig.main(args)
+
+	if args.bigwig is not None:
+		pass
+	else:
+		# Shift bam and convert to bigwig
+		import chrombpnet.helpers.preprocessing.reads_to_bigwig as reads_to_bigwig	
+		args.output_prefix = os.path.join(args.output_dir,"auxiliary/{}data".format(fpx))
+		args.plus_shift = None
+		args.minus_shift = None
+		reads_to_bigwig.main(args)
+		args.bigwig = os.path.join(args.output_dir,"auxiliary/{}data_unstranded.bw".format(fpx)) # bigwig input
 
 	# QC bigwig
 	import chrombpnet.helpers.preprocessing.analysis.build_pwm_from_bigwig as build_pwm_from_bigwig	
-	args.bigwig = os.path.join(args.output_dir,"auxiliary/{}data_unstranded.bw".format(fpx))
+
 	args.output_prefix = os.path.join(args.output_dir,"evaluation/{}bw_shift_qc".format(fpx))
 	folds = json.load(open(args.chr_fold_path))
 	assert(len(folds["valid"]) > 0) # validation list of chromosomes is empty
@@ -265,16 +269,19 @@ def train_bias_pipeline(args):
 	else:
 		fpx = ""
 
-	# Shift bam and convert to bigwig
-	import chrombpnet.helpers.preprocessing.reads_to_bigwig as reads_to_bigwig	
-	args.output_prefix = os.path.join(args.output_dir,"auxiliary/{}data".format(fpx))
-	args.plus_shift = None
-	args.minus_shift = None
-	reads_to_bigwig.main(args)
-
+	if args.bigwig is not None:
+		pass
+	else:
+		# Shift bam and convert to bigwig
+		import chrombpnet.helpers.preprocessing.reads_to_bigwig as reads_to_bigwig	
+		args.output_prefix = os.path.join(args.output_dir,"auxiliary/{}data".format(fpx))
+		args.plus_shift = None
+		args.minus_shift = None
+		reads_to_bigwig.main(args)
+		args.bigwig = os.path.join(args.output_dir,"auxiliary/{}data_unstranded.bw".format(fpx))
+
 	# QC bigwig
 	import chrombpnet.helpers.preprocessing.analysis.build_pwm_from_bigwig as build_pwm_from_bigwig	
-	args.bigwig = os.path.join(args.output_dir,"auxiliary/{}data_unstranded.bw".format(fpx))
 	args.output_prefix = os.path.join(args.output_dir,"evaluation/{}bw_shift_qc".format(fpx))
 	folds = json.load(open(args.chr_fold_path))
 	assert(len(folds["valid"]) > 0) # validation list of chromosomes is empty

diff --git a/dockerfile b/dockerfile
@@ -37,6 +37,11 @@ RUN conda install -y -c conda-forge -c bioconda samtools bedtools ucsc-bedgrapht
 
 # Install jq
 RUN apt-get install -y jq
+RUN apt-get install -y libcairo2
+RUN apt-get install -y libpango-1.0-0
+RUN apt-get install -y libpangocairo-1.0-0
+RUN apt-get install -y libgdk-pixbuf2.0-0
+RUN apt-get install -y libffi-dev
 
 # Clean up after apt and conda
 RUN apt-get clean && rm -rf /var/lib/apt/lists/*

diff --git a/requirements.txt b/requirements.txt
@@ -13,7 +13,7 @@ protobuf==3.20
 tqdm==4.48.2
 deepdish==0.3.7
 deeplift==0.6.13.0
-modisco==0.5.16.0
+modisco==0.5.14.1
 modisco-lite==2.0.5
 weasyprint==52.5
 kundajelab-shap==1