update query prompt and func

gptbioinsightor/celltype.py

@@ -12,8 +12,8 @@
 from .prompt import *
 
 
-def _query_celltype(genes, queryid, background, provider, model, base_url, sys_prompt):
-    text = CELLTYPE_PROMPT.format(setid=queryid, gene=",".join(genes), background=background)
+def _query_celltype(queryid, gene_txt, cluster_num, background, provider, model, base_url, sys_prompt):
+    text = CELLTYPE_PROMPT.format(setid=queryid, gene=gene_txt, setnum=cluster_num,background=background)
     msg = [{"role": "user", "content": text}]
     response = query_model(msg, provider=provider, model=model, base_url=base_url, sys_prompt=sys_prompt)
     return response
@@ -79,13 +79,14 @@ def get_celltype(
     if n_jobs is None:
         n_jobs = min(os.cpu_count()//2, len(gene_dic))
 
-    def _aux_func(item):
-        return _query_celltype(item[1][:topnumber], item[0], background, provider, model, base_url, sys_prompt)
+    def _aux_func(args):
+        gene_txt = "\n".join([f"cluster {k}: {','.join(genes[:topnumber])}" for k,genes in args[1].items()])
+        return _query_celltype(args[0], gene_txt, len(args[1]), background, provider, model, base_url, sys_prompt)
 
     celltype_ls = []
     with ThreadPoolExecutor(max_workers=n_jobs) as executor:
-        results = executor.map(_aux_func, gene_dic.items())
-        for (gsid, genes), res in zip(gene_dic.items(), results):
+        results = executor.map(_aux_func, [(k, gene_dic) for k in gene_dic.keys()])
+        for gsid, res in zip(gene_dic.values(), results):
             res = res.strip("```").strip("'''").strip()
             print(res, file=handle)
             ctn = ul.get_celltype_name(res)

gptbioinsightor/prompt.py

@@ -42,127 +42,63 @@
 """
 
 
-LIKELY_CELLTYPE_PROMPT = """
-Geneset {setid}:
-```gene list
+CELLTYPE_PROMPT = """
+Input:
+'''
+Geneset: 
 {gene}
-```
-
-Hi, GPTBioInsightor! Please analyze the above geneset and determine three most likely celltypes based on the following INSTRUCTIONS.
-
-INSTRUCTIONS:
-0. Evaluate each gene individually, providing evidence and rationale for every potential cell type.
-1. Prioritize cell-specific gene markers
-2. Consider context-specific gene markers and samples/cells source within context (BACKGROUND)
-3. Analyze the celltype context (BACKGROUND) to speculate on cell states, such as stress responses, invasiveness, proliferation rates, developmental stages, or other transient/dynamic properties.
-4. Focus on positive evidence; avoid using the absence of markers as primary reasoning.
-5. Evaluate the possibility of mixed cell populations or transitional states.
-6. If applicable, note any unexpected gene combinations that might suggest novel cell states or types.
 
+Context: 
+{background}. Here are {setnum} genesets for {setnum} different cell clusters up-regulated DEGs. 
+'''
 
-BACKGROUND:
-{background}
-
-Please format your output as follows, without any additional content:
+Hi, GPTBioInsightor! Please analyze Input and predict the celltypes of geneset cluster {setid} based on the following INSTRUCTIONS.
 
+INSTRUCTIONS:
+0. Analyze each gene in cluster {setid}, check cell-specific and context-specific markers
+1. prioritize single Gold Standard marker or gene marker combinations for celltype prediction
+2. Consider the context of Input for celltype prediction; e.g. tissue, disease, etc.
+3. Integrate Context of Input to speculate on some novel insights 
+4. Focus on positive evidence, avoid using marker absence as primary reasoning.
+5. Exclude celltypes with clear negative markers in the geneset.
+6. Consider each cluster has different celltype prediction in most time, exclude celltypes represented by ohter cluster gene markers. 
+7. Consider one Optimal celltypes and two alternative celltypes.
+
+Output Format:, without any additional prompt or string:
 '''
-## Geneset {setid}:
+## cluster geneset {setid}
 
 ### Gene List
 ```
-[gene list]
+[cluster {setid} gene list]
 ```
 
-### Potential Cell Types
-
-#### [CELLTYPE1]
-**Gene Markers**:
-- cell-specific: [CELL-SPECIFIC GENE MARKERS]
-- context-specific: [CONTEXT-SPECIFIC GENE MARKERS]
-
-**Evidence**: [DETAILED EVIDENCE SUPPORTING THIS CELL TYPE]
-
-**Rationale**: [COMPREHENSIVE REASONING]
-
-**Potential Cell State**: [SPECULATED STATE BASED ON BACKGROUND]
-
-#### [CELLTYPE2]
-**Gene Markers**:
-- cell-specific: [CELL-SPECIFIC GENE MARKERS]
-- context-specific: [CONTEXT-SPECIFIC GENE MARKERS]
-
-**Evidence**: [DETAILED EVIDENCE SUPPORTING THIS CELL TYPE]
-
-**Rationale**: [COMPREHENSIVE REASONING]
-
-**Potential Cell State**: [SPECULATED STATE BASED ON BACKGROUND]
-
-#### [CELLTYPE3]
-**Gene Markers**:
-- cell-specific: [CELL-SPECIFIC GENE MARKERS]
-- context-specific: [CONTEXT-SPECIFIC GENE MARKERS]
-
-**Evidence**: [DETAILED EVIDENCE SUPPORTING THIS CELL TYPE]
-
-**Rationale**: [COMPREHENSIVE REASONING]
-
-**Potential Cell State**: [SPECULATED STATE BASED ON BACKGROUND]
-
-### Additional Observations
-[ANY NOTEWORTHY PATTERNS, UNUSUAL GENE COMBINATIONS, OR POTENTIAL NOVEL INSIGHTS]
-'''
-"""
-
-
-FINAL_CELLTYPE_PROMPT = """
-Hi, GPTBioInsightor! Please determine the most likely cell type for each gene set from the provided potential cell types. Your analysis should be based on the following INSTRUCTIONS and context(BACKGROUND) information.
-
-INSTRUCTIONS:
-1. Provide comprehensive evidence and reasoning for the most likely cell type of each gene set wtih context(BACKGROUND).
-2. Prioritize cell-specific and context-specific gene markers in your analysis.
-3. Fully integrate the BACKGROUND information into your analysis to determine the most logical cell type.
-4. Speculate on the cell state, considering factors such as stress response, invasiveness, proliferation rate, developmental stage, or other transient/dynamic properties.
-5. Do not use the absence of markers as primary evidence; focus on positive evidence.
-6. Exclude cell types with clear negative markers present in the gene set.
-7. Evaluate the possibility of mixed cell populations or transitional states if the gene set suggests this.
-8. Note any unusual gene combinations or expression patterns that might indicate novel cell states or types.
-
-BACKGROUND:
-{background}. Above are {geneset_num} genesets and their potential celltypes, geneseach geneset is highly expressed relative to other gene sets..
-
-For the output you should follow this format:
-'''
-### [geneset id] : [ CELLTYPE NAME]
-
-**Gene Markers**:
-- cell-specific: [CELL-SPECIFIC GENE MARKERS]
-- context-specific: [CONTEXT-SPECIFIC GENE MARKERS]
-
-**Evidence and Reasoning**:
-1. [MAIN EVIDENCE POINT, like cell-specific marker]
-2. [SECONDARY EVIDENCE POINT, like context-specific marker ]
-3. [ADDITIONAL EVIDENCE POINTS AS NEEDED]
-
-**Cell State/Subtype**: [SPECULATED CELL STATE OR SUBTYPE BASED ON BACKGROUND]
-**Alternative Considerations**: [BRIEFLY MENTION OTHER CELL TYPES CONSIDERED AND WHY THEY WERE RULED OUT]
-
-
-### [geneset id] : [ CELLTYPE NAME ] 
+### Celltype Prediction
+#### Optimal Celltype: [OPTIMAL CELLTYPE NAME]
+**Key Markers**:
+- Cell-specific: [CELL-SPECIFIC MARKERS]
+- Context-specific: [CONTEXT-SPECIFIC MARKERS]
 
-**Gene Markers**:
-- cell-specific: [CELL-SPECIFIC GENE MARKERS]
-- context-specific: [CONTEXT-SPECIFIC GENE MARKERS]
+**Evidence and Reasoning**
+- [PRIMARY EVIDENCE]
+- [SECONDARY EVIDENCE]
+- [ADDITIONAL EVIDENCE AS NEEDED]
 
-**Evidence and Reasoning**:
-1. [MAIN EVIDENCE POINT, like cell-specific marker]
-2. [SECONDARY EVIDENCE POINT, like context-specific marker ]
-3. [ADDITIONAL EVIDENCE POINTS AS NEEDED]
+**Validation**: [OTHER Gold Standard MARKERS(NOT IN Geneset {setid}) TO VALIDATE THE OPTIMAL CELLTYPE]
 
-**Cell State/Subtype**: [SPECULATED CELL STATE OR SUBTYPE BASED ON BACKGROUND]
-**Alternative Considerations**: [BRIEFLY MENTION OTHER CELL TYPES CONSIDERED AND WHY THEY WERE RULED OUT]
+#### Alternative Considerations
+- Alternative celltype1
+    - [WHY Alternative? Key MARKERS, Evidence and Reasoning]
+    - [OTHER Gold Standard MARKERS(NOT IN Geneset {setid}) TO VALIDATE THE Alternative celltype1]
 
-[REPEAT FOR EACH GENE SET]
+- Alternative celltype2
+    - [WHY Alternative? Key MARKERS, Evidence and Reasoning]
+    - [OTHER Gold Standard MARKERS(NOT IN Geneset {setid}) TO VALIDATE THE Alternative celltype2]
 
+### Novel Insights
+- [NOTEWORTHY PATTERNS]
+- [CELL STATE]
+- [POTENTIAL NEW FINDINGS]
 '''
 """