Spaces:
Running
Running
ACMCMC
commited on
Commit
•
1e2e3b8
1
Parent(s):
dda0120
WIP
Browse files- app.py +13 -10
- clinical_trials_embeddings.ipynb +3 -3
- llm_res.py +105 -78
- requirements.txt +1 -0
app.py
CHANGED
@@ -14,6 +14,7 @@ from utils import (
|
|
14 |
get_clinical_trials_related_to_diseases,
|
15 |
get_clinical_records_by_ids
|
16 |
)
|
|
|
17 |
import json
|
18 |
import numpy as np
|
19 |
from sentence_transformers import SentenceTransformer
|
@@ -35,10 +36,6 @@ CONNECTION_STRING = f"iris://{username}:{password}@{hostname}:{port}/{namespace}
|
|
35 |
engine = create_engine(CONNECTION_STRING)
|
36 |
|
37 |
|
38 |
-
st.title("Klìnic")
|
39 |
-
st.header("", divider='rainbow')
|
40 |
-
st.text('') # dummy to add spacing
|
41 |
-
|
42 |
with st.container(): # user input
|
43 |
col1, col2 = st.columns((6, 1))
|
44 |
|
@@ -58,30 +55,36 @@ with st.container():
|
|
58 |
with st.status("Analyzing...") as status:
|
59 |
# 1. Embed the textual description that the user entered using the model
|
60 |
# 2. Get 5 diseases with the highest cosine silimarity from the DB
|
|
|
61 |
encoder = SentenceTransformer("allenai-specter")
|
62 |
diseases_related_to_the_user_text = get_diseases_related_to_a_textual_description(
|
63 |
description_input, encoder
|
64 |
)
|
65 |
-
# for disease_label in diseases_related_to_the_user_text:
|
66 |
-
# st.text(disease_label)
|
67 |
# 3. Get the similarities of the embeddings of those diseases (cosine similarity of the embeddings of the nodes of such diseases)
|
|
|
68 |
diseases_uris = [disease["uri"] for disease in diseases_related_to_the_user_text]
|
69 |
get_similarities_among_diseases_uris(diseases_uris)
|
70 |
-
#print(diseases_related_to_the_user_text)
|
71 |
# 4. Potentially filter out the diseases that are not similar enough (e.g. similarity < 0.8)
|
72 |
# 5. Augment the set of diseases: add new diseases that are similar to the ones that are already in the set, until we get 10-15 diseases
|
|
|
73 |
augmented_set_of_diseases = augment_the_set_of_diseaces(diseases_uris)
|
74 |
-
#print(augmented_set_of_diseases)
|
75 |
# 6. Query the embeddings of the diseases related to each clinical trial (also in the DB), to get the most similar clinical trials to our set of diseases
|
|
|
76 |
clinical_trials_related_to_the_diseases = get_clinical_trials_related_to_diseases(
|
77 |
augmented_set_of_diseases, encoder
|
78 |
)
|
79 |
-
|
80 |
json_of_clinical_trials = get_clinical_records_by_ids(
|
81 |
[trial["nct_id"] for trial in clinical_trials_related_to_the_diseases]
|
82 |
)
|
83 |
-
|
|
|
|
|
|
|
|
|
84 |
# 8. Use an LLM to extract numerical data from the clinical trials (e.g. number of patients, number of deaths, etc.). Get summary statistics out of that.
|
|
|
85 |
# 9. Show the results to the user: graph of the diseases chosen, summary of the clinical trials, summary statistics of the clinical trials, and list of the details of the clinical trials considered
|
86 |
status.update(label="Done!", state="complete")
|
87 |
time.sleep(1)
|
|
|
14 |
get_clinical_trials_related_to_diseases,
|
15 |
get_clinical_records_by_ids
|
16 |
)
|
17 |
+
from llm_res import process_dictionaty_with_llm_to_generate_response
|
18 |
import json
|
19 |
import numpy as np
|
20 |
from sentence_transformers import SentenceTransformer
|
|
|
36 |
engine = create_engine(CONNECTION_STRING)
|
37 |
|
38 |
|
|
|
|
|
|
|
|
|
39 |
with st.container(): # user input
|
40 |
col1, col2 = st.columns((6, 1))
|
41 |
|
|
|
55 |
with st.status("Analyzing...") as status:
|
56 |
# 1. Embed the textual description that the user entered using the model
|
57 |
# 2. Get 5 diseases with the highest cosine silimarity from the DB
|
58 |
+
status.write("Analyzing the description that you wrote...")
|
59 |
encoder = SentenceTransformer("allenai-specter")
|
60 |
diseases_related_to_the_user_text = get_diseases_related_to_a_textual_description(
|
61 |
description_input, encoder
|
62 |
)
|
|
|
|
|
63 |
# 3. Get the similarities of the embeddings of those diseases (cosine similarity of the embeddings of the nodes of such diseases)
|
64 |
+
status.write("Getting the similarities among the diseases to filter out less promising ones...")
|
65 |
diseases_uris = [disease["uri"] for disease in diseases_related_to_the_user_text]
|
66 |
get_similarities_among_diseases_uris(diseases_uris)
|
|
|
67 |
# 4. Potentially filter out the diseases that are not similar enough (e.g. similarity < 0.8)
|
68 |
# 5. Augment the set of diseases: add new diseases that are similar to the ones that are already in the set, until we get 10-15 diseases
|
69 |
+
status.write("Augmenting the set of diseases by finding others with related embeddings...")
|
70 |
augmented_set_of_diseases = augment_the_set_of_diseaces(diseases_uris)
|
71 |
+
# print(augmented_set_of_diseases)
|
72 |
# 6. Query the embeddings of the diseases related to each clinical trial (also in the DB), to get the most similar clinical trials to our set of diseases
|
73 |
+
status.write("Getting the clinical trials related to the diseases found...")
|
74 |
clinical_trials_related_to_the_diseases = get_clinical_trials_related_to_diseases(
|
75 |
augmented_set_of_diseases, encoder
|
76 |
)
|
77 |
+
status.write("Getting the details of the clinical trials...")
|
78 |
json_of_clinical_trials = get_clinical_records_by_ids(
|
79 |
[trial["nct_id"] for trial in clinical_trials_related_to_the_diseases]
|
80 |
)
|
81 |
+
status.json(json_of_clinical_trials)
|
82 |
+
# 7. Use an LLM to get a summary of the clinical trials, in plain text format.
|
83 |
+
status.write("Getting a summary of the clinical trials...")
|
84 |
+
response = process_dictionaty_with_llm_to_generate_response(json_of_clinical_trials)
|
85 |
+
print(f'Response from LLM: {response}')
|
86 |
# 8. Use an LLM to extract numerical data from the clinical trials (e.g. number of patients, number of deaths, etc.). Get summary statistics out of that.
|
87 |
+
status.write("Getting summary statistics of the clinical trials...")
|
88 |
# 9. Show the results to the user: graph of the diseases chosen, summary of the clinical trials, summary statistics of the clinical trials, and list of the details of the clinical trials considered
|
89 |
status.update(label="Done!", state="complete")
|
90 |
time.sleep(1)
|
clinical_trials_embeddings.ipynb
CHANGED
@@ -61,9 +61,9 @@
|
|
61 |
"metadata": {},
|
62 |
"outputs": [],
|
63 |
"source": [
|
64 |
-
"
|
65 |
-
"
|
66 |
-
")"
|
67 |
]
|
68 |
},
|
69 |
{
|
|
|
61 |
"metadata": {},
|
62 |
"outputs": [],
|
63 |
"source": [
|
64 |
+
"from dotenv import load_dotenv\n",
|
65 |
+
"\n",
|
66 |
+
"load_dotenv()"
|
67 |
]
|
68 |
},
|
69 |
{
|
llm_res.py
CHANGED
@@ -21,6 +21,10 @@ from langchain_core.pydantic_v1 import BaseModel, Field
|
|
21 |
from langchain_openai import ChatOpenAI
|
22 |
from typing import List, Dict, Any
|
23 |
import requests
|
|
|
|
|
|
|
|
|
24 |
|
25 |
# getting the json files
|
26 |
def get_clinical_record_info(clinical_record_id: str) -> Dict[str, Any]:
|
@@ -31,6 +35,7 @@ def get_clinical_record_info(clinical_record_id: str) -> Dict[str, Any]:
|
|
31 |
response = requests.get(request_url, headers={"accept": "application/json"})
|
32 |
return response.json()
|
33 |
|
|
|
34 |
def get_clinical_records_by_ids(clinical_record_ids: List[str]) -> List[Dict[str, Any]]:
|
35 |
clinical_records = []
|
36 |
for clinical_record_id in clinical_record_ids:
|
@@ -38,80 +43,99 @@ def get_clinical_records_by_ids(clinical_record_ids: List[str]) -> List[Dict[str
|
|
38 |
clinical_records.append(clinical_record_info)
|
39 |
return clinical_records
|
40 |
|
41 |
-
|
42 |
-
|
43 |
-
# Open the JSON file for reading
|
44 |
-
with open(json_file, 'r') as f:
|
45 |
-
data = json.load(f) # Parse JSON data into a Python dictionary
|
46 |
|
47 |
# Define the fields you want to keep
|
48 |
-
fields_to_keep = [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
49 |
|
50 |
# Iterate through the dictionary and keep only the desired fields
|
51 |
filtered_data = []
|
52 |
for item in data:
|
53 |
try:
|
54 |
-
organization_name= item[
|
|
|
|
|
55 |
except:
|
56 |
-
organization_name= ""
|
57 |
try:
|
58 |
-
project_title= item[
|
|
|
|
|
59 |
except:
|
60 |
-
project_title= ""
|
61 |
-
try:
|
62 |
-
status= item[
|
63 |
except:
|
64 |
-
status= ""
|
65 |
try:
|
66 |
-
brief_description= item[
|
|
|
|
|
67 |
except:
|
68 |
-
brief_description= ""
|
69 |
try:
|
70 |
-
detailed_description= item[
|
|
|
|
|
71 |
except:
|
72 |
-
detailed_description= ""
|
73 |
try:
|
74 |
-
conditions= item[
|
75 |
except:
|
76 |
-
conditions= []
|
77 |
try:
|
78 |
-
keywords= item[
|
79 |
except:
|
80 |
-
keywords= []
|
81 |
try:
|
82 |
-
interventions= item[
|
|
|
|
|
83 |
except:
|
84 |
-
interventions= []
|
85 |
try:
|
86 |
-
primary_outcomes= item[
|
|
|
|
|
87 |
except:
|
88 |
-
primary_outcomes= []
|
89 |
try:
|
90 |
-
secondary_outcomes= item[
|
|
|
|
|
91 |
except:
|
92 |
-
secondary_outcomes= []
|
93 |
try:
|
94 |
-
eligibility= item[
|
95 |
except:
|
96 |
-
eligibility= {}
|
97 |
-
filtered_item = {
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
|
|
|
|
107 |
filtered_data.append(filtered_item)
|
108 |
|
109 |
# for ele in filtered_data:
|
110 |
# print(ele)
|
111 |
|
112 |
-
# Write the filtered data to a new JSON file
|
113 |
-
with open('output.json', 'w') as f:
|
114 |
-
json.dump(filtered_data, f, indent=4)
|
115 |
|
116 |
def llm_config():
|
117 |
tagging_prompt = ChatPromptTemplate.from_template(
|
@@ -127,20 +151,38 @@ def llm_config():
|
|
127 |
)
|
128 |
|
129 |
class Classification(BaseModel):
|
130 |
-
description: str = Field(
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
136 |
# secondary_outcomes: list= Field(description= "get the secondary outcomes of each clinical trial")
|
137 |
-
eligibility: list= Field(
|
|
|
|
|
138 |
# healthy_volunteers: list= Field(description= "determine whether the clinical trial requires healthy volunteers")
|
139 |
-
minimum_age: list = Field(
|
140 |
-
|
|
|
|
|
|
|
|
|
141 |
gender: list = Field(description="get the gender from each experiment")
|
142 |
|
143 |
-
def get_dict(self):
|
144 |
return {
|
145 |
"summary": self.description,
|
146 |
"project_title": self.project_title,
|
@@ -153,45 +195,30 @@ def llm_config():
|
|
153 |
# "healthy_volunteers": self.healthy_volunteers,
|
154 |
"minimum_age": self.minimum_age,
|
155 |
"maximum_age": self.maximum_age,
|
156 |
-
"gender": self.gender
|
157 |
}
|
158 |
-
|
159 |
# LLM
|
160 |
llm = ChatOpenAI(
|
161 |
-
temperature=0.6,
|
162 |
model="gpt-4",
|
163 |
-
openai_api_key="
|
164 |
-
).with_structured_output(
|
165 |
-
Classification
|
166 |
-
)
|
167 |
|
168 |
tagging_chain = tagging_prompt | llm
|
169 |
-
|
170 |
return tagging_chain
|
171 |
|
172 |
-
def get_llm_results(results):
|
173 |
-
result_dict= results.get_dict()
|
174 |
-
return result_dict
|
175 |
|
176 |
-
def save_llm_results(results_json):
|
177 |
-
with open('llm_results.json', 'w') as f:
|
178 |
-
json.dump(results_json, f, indent=4)
|
179 |
-
|
180 |
# clinical_record_info = get_clinical_records_by_ids(['NCT00841061', 'NCT03035123', 'NCT02272751', 'NCT03035123', 'NCT03055377'])
|
181 |
# print(clinical_record_info)
|
182 |
|
183 |
# with open('data.json', 'w') as f:
|
184 |
# json.dump(clinical_record_info, f, indent=4)
|
185 |
|
186 |
-
|
187 |
-
json_file= "D:/HACKUPC/hupc/klinic/data.json"
|
188 |
-
process_json(json_file)
|
189 |
-
|
190 |
-
with open('output.json', 'r') as file:
|
191 |
-
data = json.load(file)
|
192 |
|
193 |
-
|
194 |
-
|
195 |
-
|
196 |
-
|
197 |
-
print(result_json)
|
|
|
21 |
from langchain_openai import ChatOpenAI
|
22 |
from typing import List, Dict, Any
|
23 |
import requests
|
24 |
+
from dotenv import load_dotenv
|
25 |
+
|
26 |
+
load_dotenv()
|
27 |
+
|
28 |
|
29 |
# getting the json files
|
30 |
def get_clinical_record_info(clinical_record_id: str) -> Dict[str, Any]:
|
|
|
35 |
response = requests.get(request_url, headers={"accept": "application/json"})
|
36 |
return response.json()
|
37 |
|
38 |
+
|
39 |
def get_clinical_records_by_ids(clinical_record_ids: List[str]) -> List[Dict[str, Any]]:
|
40 |
clinical_records = []
|
41 |
for clinical_record_id in clinical_record_ids:
|
|
|
43 |
clinical_records.append(clinical_record_info)
|
44 |
return clinical_records
|
45 |
|
46 |
+
|
47 |
+
def process_json_data_for_llm(data):
|
|
|
|
|
|
|
48 |
|
49 |
# Define the fields you want to keep
|
50 |
+
fields_to_keep = [
|
51 |
+
"class_of_organization",
|
52 |
+
"title",
|
53 |
+
"overallStatus",
|
54 |
+
"descriptionModule",
|
55 |
+
"conditions",
|
56 |
+
"interventions",
|
57 |
+
"outcomesModule",
|
58 |
+
"eligibilityModule",
|
59 |
+
]
|
60 |
|
61 |
# Iterate through the dictionary and keep only the desired fields
|
62 |
filtered_data = []
|
63 |
for item in data:
|
64 |
try:
|
65 |
+
organization_name = item["protocolSection"]["identificationModule"][
|
66 |
+
"organization"
|
67 |
+
]["fullName"]
|
68 |
except:
|
69 |
+
organization_name = ""
|
70 |
try:
|
71 |
+
project_title = item["protocolSection"]["identificationModule"][
|
72 |
+
"officialTitle"
|
73 |
+
]
|
74 |
except:
|
75 |
+
project_title = ""
|
76 |
+
try:
|
77 |
+
status = item["protocolSection"]["statusModule"]["overallStatus"]
|
78 |
except:
|
79 |
+
status = ""
|
80 |
try:
|
81 |
+
brief_description = item["protocolSection"]["descriptionModule"][
|
82 |
+
"briefSummary"
|
83 |
+
]
|
84 |
except:
|
85 |
+
brief_description = ""
|
86 |
try:
|
87 |
+
detailed_description = item["protocolSection"]["descriptionModule"][
|
88 |
+
"detailedDescription"
|
89 |
+
]
|
90 |
except:
|
91 |
+
detailed_description = ""
|
92 |
try:
|
93 |
+
conditions = item["protocolSection"]["conditionsModule"]["conditions"]
|
94 |
except:
|
95 |
+
conditions = []
|
96 |
try:
|
97 |
+
keywords = item["protocolSection"]["conditionsModule"]["keywords"]
|
98 |
except:
|
99 |
+
keywords = []
|
100 |
try:
|
101 |
+
interventions = item["protocolSection"]["armsInterventionsModule"][
|
102 |
+
"interventions"
|
103 |
+
]
|
104 |
except:
|
105 |
+
interventions = []
|
106 |
try:
|
107 |
+
primary_outcomes = item["protocolSection"]["outcomesModule"][
|
108 |
+
"primaryOutcomes"
|
109 |
+
]
|
110 |
except:
|
111 |
+
primary_outcomes = []
|
112 |
try:
|
113 |
+
secondary_outcomes = item["protocolSection"]["outcomesModule"][
|
114 |
+
"secondaryOutcomes"
|
115 |
+
]
|
116 |
except:
|
117 |
+
secondary_outcomes = []
|
118 |
try:
|
119 |
+
eligibility = item["protocolSection"]["eligibilityModule"]
|
120 |
except:
|
121 |
+
eligibility = {}
|
122 |
+
filtered_item = {
|
123 |
+
"organization_name": organization_name,
|
124 |
+
"project_title": project_title,
|
125 |
+
"status": status,
|
126 |
+
"brief_description": brief_description,
|
127 |
+
"detailed_description": detailed_description,
|
128 |
+
"keywords": keywords,
|
129 |
+
"interventions": interventions,
|
130 |
+
"primary_outcomes": primary_outcomes,
|
131 |
+
"secondary_outcomes": secondary_outcomes,
|
132 |
+
"eligibility": eligibility,
|
133 |
+
}
|
134 |
filtered_data.append(filtered_item)
|
135 |
|
136 |
# for ele in filtered_data:
|
137 |
# print(ele)
|
138 |
|
|
|
|
|
|
|
139 |
|
140 |
def llm_config():
|
141 |
tagging_prompt = ChatPromptTemplate.from_template(
|
|
|
151 |
)
|
152 |
|
153 |
class Classification(BaseModel):
|
154 |
+
description: str = Field(
|
155 |
+
description="text description grouping all the clinical trials using brief_description and detailed_description keys"
|
156 |
+
)
|
157 |
+
project_title: list = Field(
|
158 |
+
description="Extract the project title of all the clinical trials"
|
159 |
+
)
|
160 |
+
status: list = Field(
|
161 |
+
description="Extract the status of all the clinical trials"
|
162 |
+
)
|
163 |
+
keywords: list = Field(
|
164 |
+
description="Extract the most relevant keywords regrouping all the clinical trials"
|
165 |
+
)
|
166 |
+
interventions: list = Field(
|
167 |
+
description="describe the interventions for each clinical trial using title, name and description"
|
168 |
+
)
|
169 |
+
primary_outcomes: list = Field(
|
170 |
+
description="get the primary outcomes of each clinical trial"
|
171 |
+
)
|
172 |
# secondary_outcomes: list= Field(description= "get the secondary outcomes of each clinical trial")
|
173 |
+
eligibility: list = Field(
|
174 |
+
description="get the eligibilityCriteria grouping all the clinical trials"
|
175 |
+
)
|
176 |
# healthy_volunteers: list= Field(description= "determine whether the clinical trial requires healthy volunteers")
|
177 |
+
minimum_age: list = Field(
|
178 |
+
description="get the minimum age from each experiment"
|
179 |
+
)
|
180 |
+
maximum_age: list = Field(
|
181 |
+
description="get the maximum age from each experiment"
|
182 |
+
)
|
183 |
gender: list = Field(description="get the gender from each experiment")
|
184 |
|
185 |
+
def get_dict(self):
|
186 |
return {
|
187 |
"summary": self.description,
|
188 |
"project_title": self.project_title,
|
|
|
195 |
# "healthy_volunteers": self.healthy_volunteers,
|
196 |
"minimum_age": self.minimum_age,
|
197 |
"maximum_age": self.maximum_age,
|
198 |
+
"gender": self.gender,
|
199 |
}
|
200 |
+
|
201 |
# LLM
|
202 |
llm = ChatOpenAI(
|
203 |
+
temperature=0.6,
|
204 |
model="gpt-4",
|
205 |
+
openai_api_key=os.environ["OPENAI_API_KEY"],
|
206 |
+
).with_structured_output(Classification)
|
|
|
|
|
207 |
|
208 |
tagging_chain = tagging_prompt | llm
|
209 |
+
|
210 |
return tagging_chain
|
211 |
|
|
|
|
|
|
|
212 |
|
|
|
|
|
|
|
|
|
213 |
# clinical_record_info = get_clinical_records_by_ids(['NCT00841061', 'NCT03035123', 'NCT02272751', 'NCT03035123', 'NCT03055377'])
|
214 |
# print(clinical_record_info)
|
215 |
|
216 |
# with open('data.json', 'w') as f:
|
217 |
# json.dump(clinical_record_info, f, indent=4)
|
218 |
|
219 |
+
tagging_chain = llm_config()
|
|
|
|
|
|
|
|
|
|
|
220 |
|
221 |
+
def process_dictionaty_with_llm_to_generate_response(json_contents):
|
222 |
+
processed_data = process_json_data_for_llm(json_contents)
|
223 |
+
res = tagging_chain.invoke({"input": processed_data})
|
224 |
+
return res
|
|
requirements.txt
CHANGED
@@ -10,3 +10,4 @@ openai==1.25.1
|
|
10 |
sentence_transformers==2.7.0
|
11 |
streamlit-agraph
|
12 |
streamlit==1.34.0
|
|
|
|
10 |
sentence_transformers==2.7.0
|
11 |
streamlit-agraph
|
12 |
streamlit==1.34.0
|
13 |
+
langchain-openai==0.1.6
|