Fast-HInt-ppi 0.2.0__tar.gz → 0.2.2__tar.gz

{fast_hint_ppi-0.2.0 → fast_hint_ppi-0.2.2}/Fast_HInt/Fast_HInt.py

@@ -187,7 +187,7 @@ def main() :
     create_feature(HInt_object, Informations_dict, GPU, CPU, first_need_msa, first_need_pkl)
 
     HInt_object.Make_save_dict()
-    prio_list_MSA = prioritize_by_vram_fit(job_with_vram_length, GPU)
+    prio_list_MSA = prioritize_by_vram_fit(job_with_vram_length, First_batch, GPU)
     batch_MSA = []
     if prio_list_MSA != [] :
         prio_list_MSA.pop(0) #remove first batch
@@ -198,7 +198,8 @@ def main() :
                     new_prio_list.append(prot)
                 if prot in need_pkl :
                     new_prio_list.append(prot)
-            batch_MSA.append(new_prio_list)
+            if new_prio_list != [] :
+                batch_MSA.append(new_prio_list)
 
     if len(first_need_msa) > 0 :
         logger.info("Generating MSA depth figures")

{fast_hint_ppi-0.2.0 → fast_hint_ppi-0.2.2}/Fast_HInt/Scoring_HInt.py

@@ -24,8 +24,6 @@ import shutil
 from pathlib import Path
 from multiprocessing import Pool
 from tqdm import tqdm
-from datetime import datetime
-from scipy.special import softmax
 from Bio import PDB
 
 # Configure global logger
@@ -94,7 +92,6 @@ def Score_interaction (file, Informations_dict, CPU, Interaction, job="", multi_
                 bait_name = bait_name.replace(prot,f"{prot}_{start}-{end}")
         possible_prey = file.get_possible_prey()
     if job != "" :
-        print("why")
         bait = job.replace(job.split(";")[-1],"").strip(";")
         bait_name = bait.replace(";","_and_")
         possible_prey = [job.split(";")[-1].strip("\n")]

{fast_hint_ppi-0.2.0 → fast_hint_ppi-0.2.2}/Fast_HInt/Utils_HInt.py

@@ -90,7 +90,7 @@ def Define_informations() :
                 logger.info("Set pickle feature path by default on ./feature")
                 Informations_dict[informations_key] = "./feature"
             elif informations_key == "Path_MMseqs2_Data" :
-                logger.info("/!\ local MMseqs2 GPU will not be used")
+                logger.info("Warning : local MMseqs2 GPU will not be used")
             elif informations_key == "Signal_peptide" :
                 Informations_dict[informations_key] = "None"
             elif informations_key == "Homo-oligomer" :
@@ -454,21 +454,19 @@ def create_feature (file, Informations_dict, GPU, CPU, need_msa, need_pkl) :
         max_workers_mafft = CPU # how many jobs in parallel
 
         #Look for MSA in AlphaFold database
-        logger.info(f"Search MSA in AlphaFold database")
+        logger.info(f"Searching for MSAs in the AlphaFold database")
 
         retry_queue = AFDB_msa_search(Path_Pickle_Feature,prot_SP,prot_no_SP,generated_msa)
 
     while retry_queue != [] : #retry if error during MSA search in AFdb
         retry_queue = AFDB_msa_search(Path_Pickle_Feature,prot_SP,prot_no_SP,retry_queue)
 
-    logger.info("MSA search in AlphaFold database complete")     
+    logger.info("MSA search in the AlphaFold database completed")     
     file.create_fasta_file(False, need_msa, need_pkl)
 
     #Create MSA files with ColabFold mmseq2 classic pipeline for proteins without MSA, and pickle files for proteins generated with MMseqs2 GPU
-    if AF_version == "2" : # Cause AF3 if too fast for hmmer MSA generation, so we use mmseqs2 classic pipeline for AF3
-        mmseqs2 = "True"
-    else :
-        mmseqs2 = "False"
+    # Cause AF3 if too fast for hmmer MSA generation, so we use mmseqs2 classic pipeline for AF3
+
     if len(need_msa) > 0 :
         max_workers_feature = CPU
         futures_list = []
@@ -477,17 +475,17 @@ def create_feature (file, Informations_dict, GPU, CPU, need_msa, need_pkl) :
                 msa_name = f"./log_file/{protein}_msa.fasta"
                 cmd = ["create_individual_features.py",
                 f"--fasta_paths={msa_name}",
-                f"--data_dir=/data/alphadata_v2",
+                f"--data_dir={Path_AlphaFold_Data}",
                 "--save_msa_files=True",
                 f"--output_dir={Path_Pickle_Feature}",
                 "--max_template_date=2024-05-02",
                 "--skip_existing=True",
-                f"--use_mmseqs2={mmseqs2}",
+                f"--use_mmseqs2=True",
                 "--use_precomputed_msas=True"]
                 futures_list.append(executor.submit(create_ind_feature, protein, msa_name, prot_no_SP, cmd))
     end = time.time()
     elapsed = end - start
-    logger.info("Create MSA take "+ str(elapsed/60)+" minutes")
+    logger.info("MSA creation took "+ str(elapsed/60)+" minutes")
 
     #Create pkl files for proteins without pkl file (from MSA found in AFdb or error during MSA generation with ColabFold mmseqs2)
     if len(need_pkl) > 0 :
@@ -972,46 +970,82 @@ def Generate_first_batch(job_with_vram_length, GPU, multi_job_per_gpu) :
 
     return gpu_jobs, gpu_vram_used
 
-def prioritize_by_vram_fit(jobs, GPU, max_per_batch=3):
-
-    import pynvml
+def prioritize_by_vram_fit(jobs, first_batch, GPU, max_per_batch=3) :
 
     pynvml.nvmlInit()
     handle = pynvml.nvmlDeviceGetHandleByIndex(int(GPU[0]))
     max_vram = pynvml.nvmlDeviceGetMemoryInfo(handle).total / 1024**3
     pynvml.nvmlShutdown()
 
-    jobs = sorted(jobs, key=lambda x: x[1], reverse=True)
+    jobs = [(job.strip().split(";")[1], vram) for job, vram in jobs]
+
+    jobs_dict = dict(jobs)
+
+    current_batch = [
+        (job, jobs_dict[job])
+        for job in first_batch[0]
+        if job in jobs_dict
+    ]
+
+    current_jobs = [job for job, _ in current_batch]
+
+    remaining = [(job, vram) for job, vram in jobs if job not in current_jobs]
 
-    remaining = jobs.copy()
     all_batches = []
+    pending = []
+    alternate = 1
+
+    while remaining :
+        current_batch.sort(key=lambda x: x[1]) #small finish first
 
-    while remaining:
+        finished_job, finished_vram = current_batch.pop(0)
 
+        total_vram = sum(vram for _, vram in current_batch)
         batch = []
-        used_vram = 0.0
 
-        # on tente plusieurs passes pour remplir correctement
-        changed = True
+        remaining.sort(
+            key=lambda x: x[1],
+            reverse=True
+        )
 
-        while changed and len(batch) < max_per_batch:
-            changed = False
+        while len(batch) < max_per_batch :
+            added = False
+            for job, vram in remaining :
+                if total_vram + vram <= max_vram or (vram <= finished_vram and alternate % 2 == 0) :
+                    alternate += 1
+                    batch.append(job)
+                    total_vram += vram
+                    added = True
+                    break
+                    
 
-            for i in range(len(remaining)):
+            if not added :
+                break
 
-                job, vram = remaining[i]
+        if not batch :
+            continue
 
-                if len(batch) < max_per_batch and used_vram + vram <= max_vram:
-                    batch.append(job.split(";")[-1].strip("\n"))
-                    used_vram += vram
-                    remaining.pop(i)
-                    changed = True
-                    break
+        remaining = [
+            (job, vram)
+            for job, vram in remaining
+            if job not in batch
+        ]
 
-        all_batches.append(batch)
+        for job in batch :
+            current_batch.append((job, jobs_dict[job]))
 
-    return all_batches
+        for job in batch :
+            if job not in pending :
+                pending.append(job)
+
+        while len(pending) >= max_per_batch :
+            all_batches.append(pending[:max_per_batch])
+            pending = pending[max_per_batch:]
 
+    if pending :
+        all_batches.append(pending)
+
+    return all_batches
 def Generate_3D_model(HInt_object, CPU, multi_scoring, Informations_dict, interaction_type, job_with_vram_length, GPU, multi_job_per_gpu, seq_bait={}, compounds_dict={}) :
     """
     Generate 3D models using multiple GPUs and multiprocessing.

{fast_hint_ppi-0.2.0 → fast_hint_ppi-0.2.2}/Fast_HInt_ppi.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: Fast_HInt-ppi
-Version: 0.2.0
+Version: 0.2.2
 Summary: A tool to find homologous interactions and speed up AlphaFold-based structural modeling.
 Home-page: https://github.com/Qrouger/HInt
 Author: Quentin Rouger

{fast_hint_ppi-0.2.0 → fast_hint_ppi-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: Fast_HInt-ppi
-Version: 0.2.0
+Version: 0.2.2
 Summary: A tool to find homologous interactions and speed up AlphaFold-based structural modeling.
 Home-page: https://github.com/Qrouger/HInt
 Author: Quentin Rouger

{fast_hint_ppi-0.2.0 → fast_hint_ppi-0.2.2}/setup.py

@@ -4,7 +4,7 @@ from setuptools import setup
 
 setup(
     name='Fast_HInt-ppi',
-    version='0.2.0',
+    version='0.2.2',
     description=(
         'A tool to find homologous interactions and speed up AlphaFold-based structural modeling.'
     ),