CTInfer
Compound Target Inference Tool

Technical Documentation, User Guide & GDPR Compliance

Version	1.0
Date	30 June 2026
Author	Maxence Belin
Status	Final
License	GNU GENERAL PUBLIC LICENSE Version 3, 29 June 2007

1. Overview

CTInfer (Compound Target Inference Tool) is a desktop application developed for IARC to automatically identify proteins inhibited by chemical compounds. Starting from a list of compounds and their PubChem identifiers (CID), CTInfer simultaneously queries tree reference databases, PubChem, MedChemExpress and canSAR.ai, extracts molecular targets, detection keywords, biological descriptions and bibliographic references, then generates a structured Excel file.

CTInfer is designed to process hundreds of compounds in a single session, combining the intelligence of three complementary sources: PubChem (NCBI public database) and MedChemExpress (supplier of bioactive compounds) and canSAR.ai (oncology database of the MD Anderson Cancer Center).

1.1 Problem Solved

Manually identifying molecular targets for a panel of compounds is a lengthy and repetitive task. For each compound, it is necessary to consult PubChem, MedChemExpress, and canSAR.ia, extract the relevant information, consolidate it, and format it. For more than 500 compounds, this represents hours of work.

CTInfer fully automates this process and generates a structured Excel spreadsheet in just a few dozen minutes.

1.2 Main capabilities

• Load an Excel or CSV file containing compound names and their PubChem CIDs

• Automatic query of MedChemExpress via a real browser (Playwright/Chromium)

• Query PubChem via the PUG View API (3 cascading endpoints)

• Extraction of inhibited molecular targets from MCE /Targets/ links

• Target extraction using 14+ different regex patterns on description texts

• Retrieval of complete biological descriptions (MCE and PubChem)

• Retrieval of all bibliographic references (MCE: References tab; PubChem: Reference fields)

• Generation of a structured Excel file with conditional columns (only active sources appear)

• MCE search fallback if the direct URL does not work

• Multilingual interface: French, English, Spanish, Portuguese, German, Chinese

• Real-time log with per-compound progress

• Query of canSAR.ai (target affinity): retrieves validated Homo sapiens targets with Mean Potency < 1000 nM

2. Installation

2.1 Requirements

Requirement	Details
Operating system	Windows 10 or 11 (64-bit)
Python	Version 3.8 or higher
Internet connection	Required to query PubChem and MCE
Disk space	~500 MB (Chromium included)

2.2 CTInfer folder contents

File	Role
CTInfer_app.py	Main application script
translations.py	Translations (6 languages)
Lancer_CTInfer.bat	Windows launcher installs dependencies automatically
CTInfer.ico	Application icon
settings.json	User preferences (created automatically)

2.3 Installation steps

Step 1: Install Python

• Go to https://python.org/downloads

• Download the latest stable version (3.11 or higher recommended)

• During installation, make sure to check the Add Python to PATH box

Step 2: First launch

• Double-click Lancer_CTInfer.bat

• Dependencies are installed automatically: Playwright, Chromium, openpyxl, beautifulsoup4

• This installation only occurs once (approximately 3–5 minutes)

Do not close the black window during the automatic Chromium installation.

3. Obtaining PubChem CIDs

CTInfer requires PubChem CID identifiers for each compound. If you only have molecule names, here is the procedure to obtain the corresponding CIDs in just a few minutes.

The CID (Compound ID) is the unique numeric identifier assigned by PubChem to each molecule. It is required for CTInfer to query the database.

3.1 Converting names to CIDs via PubChem ID Exchange

PubChem offers a free online tool to convert a bulk list of molecule names into CIDs. Here is the procedure:

1	Go to the PubChem ID Exchange service: https://pubchem.ncbi.nlm.nih.gov/idexchange/

2	In the Input ID List section, select "Synonyms" from the dropdown menu this allows you to enter common names, IUPAC names, trade names, CAS numbers, etc.

3	Paste the list of molecule names directly into the text field (one name per line), or click Import File to load a .txt file containing the list.

4	In the Output IDs section, select "CIDs" from the dropdown menu.

5	Click Submit Job and wait a few seconds. The corresponding CIDs are displayed or can be downloaded.

3.2 File format for CTInfer

Once the CIDs are obtained, prepare an Excel file with at least two columns:

Column	Content	Example
Compound	Compound name	Dacinostat
CID	Numeric PubChem identifier	6445533

If a compound is not found by PubChem ID Exchange, try name variants (trade name, CAS number, IUPAC name). Some very recent or proprietary compounds may not be listed in PubChem.

4. Usage

4.1 Preparing the source file

CTInfer automatically detects the CID column (whose name contains "cid") and the compound column (whose name contains "compound").

Column	Content	Example
Compound (or similar)	Compound name	Sangivamycin
CID (or similar)	Numeric PubChem identifier	14978

CTInfer automatically detects the CID column (whose name contains "cid") and the compound column (whose name contains "compound").

4.2 Choosing data sources

In the "3. Data sources" section of the left panel, two checkboxes allow you to select which databases to query:

Option	Behaviour
☑ MCE + ☑ PubChem (default)	Queries both sources and merges results
☑ MCE only	Searches MedChemExpress only (faster)
☑ PubChem only	Searches via PubChem API only (faster)
☑ CanSAR	Additionally queries the canSAR.ai database (target affinity); only keeps proteins where the organism is Homo sapiens and the Mean Potency is strictly below 1000 nM. Unchecked by default (slower, search-based navigation).

4.3 Running a search

Step	Action
1	Load the Excel or CSV file
2	Select the CID column from the dropdown menu
3	Choose the output folder
4	Check the desired sources (MCE, PubChem, or both)
5	Click Run search
6	Monitor progress in the real-time log
7	Retrieve the Results_targets.xlsx file in the chosen folder (a _1, _2… suffix is added automatically if a file with the same name already exists)

4.4 Words to exclude

The "Words to exclude" field contains terms that, if they precede an inhibitor in the text, invalidate the detection. For example, the word "not" in "does not inhibit X" prevents X from being listed as a target. The default values (not, no) are suitable for the majority of cases.

4.5 Estimated duration

Volume	MCE + PubChem	MCE only	PubChem only
30 compounds	5–10 min	4–8 min	1–2 min
100 compounds	20–35 min	15–25 min	3–6 min
350 compounds	60–90 min	60–90 min	10–20 min

The main time cost is linked to MedChemExpress (Playwright loads each page in 3–5 seconds like a real browser). PubChem is queried via API and is faster.

5. Excel output file

5.1 File structure

The Results_targets.xlsx file contains columns with automatic filter and fixed header row. Only columns for active sources appear:

Column	Content
Compound Name	Compound name
CID	PubChem identifier
Target(s)	Inhibited proteins, separated by semicolons
Targets CanSAR	Proteins from canSAR.ai (Homo sapiens, Mean Potency < 1000 nM), if the CanSAR source is checked, empty otherwise
Keyword MCE	Word(s) that triggered detection from MCE
Keyword PubChem	Word(s) that triggered detection from PubChem
Description MCE	Pharmacological sentence extracted from MedChemExpress
Description PubChem	Pharmacological sentence from PubChem (empty if no pharmacological keyword found)
References MCE	MCE bibliographic references
References PubChem	PubChem references

5.2 Formatting

• Green rows: compound with at least one target found

• Yellow rows: compound with no target identified

• Automatic filter on all columns

• Fixed header row (freeze panes)

6. Target extraction mechanism

6.1 Source priority

Priority	Source	Method
1	MCE	/Targets/ links on the MCE product page
2	MCE regex	14+ regex patterns on the MCE description text
3	PubChem regex	14+ regex patterns on the PubChem text
4	CanSAR	Searches the compound by name on canSAR.ai, reads the Target Affinity table, strict filter Homo sapiens + Mean Potency < 1000 nM. Does not follow the MCE/PubChem regex patterns.

6.2 Detection patterns

Pattern	Example
MCE /Targets/ links	ERK
"inhibitor of X (ABR)"	inhibitor of histone deacetylase (HDAC) → HDAC
"X inhibitor"	CDK9 inhibitor → CDK9
"inhibits X"	inhibits PKC → PKC
"designed to inhibit X"	designed to inhibit IGF1R → IGF1R
"dual X and Y inhibitor"	dual PI3K and HDAC inhibitor → PI3K, HDAC
"inhibition of X pathway"	inhibition of PI3K/Akt/mTOR pathway → PI3K/Akt/mTOR
"suppression of X"	suppression of VEGFR-2 signaling → VEGFR-2
"X/Y/Z pathway"	PI3K/Akt/mTOR pathway → PI3K/Akt/mTOR
"through inhibition of X"	through inhibition of Akt → Akt
"blocks X"	blocks ATM pathway → ATM
"binds to / targets X"	binds to Aurora kinases → Aurora kinases
"X kinase/receptor inhibitor"	Aurora B kinase inhibitor → Aurora B
"EC code (full name) inhibitor"	EC 2.7.11.13 (protein kinase C) inhibitor → protein kinase C

6.3 Anti-false-positive filters

• Noise words at the start of a target: a, an, the, both, many, novel, potent, selective, cancer, tumor, cell, activity...

• Generic words alone: inhibitor, kinase, receptor, enzyme...

• Cell lines: HepG2, MCF7, HCC, X/Y patterns like G2/ADM

• Sentences with verbs: is, are, was, were, has, have, can, will...

• User-configurable exclude words (not, no by default)

6.4 Special case: full name + acronym

When a sentence contains the full name followed by the acronym in parentheses, CTInfer extracts the acronym in priority. Example: "histone deacetylase (HDAC) inhibitor" → HDAC.

7. Technical architecture

7.1 Technologies and libraries

Library	Version	Role	License
Python	3.8+	Main language	PSF
tkinter	stdlib	Graphical interface	PSF (included)
Playwright	≥ 1.40	MCE navigation via Chromium	Apache 2.0
BeautifulSoup4	≥ 4.12	HTML parsing	MIT
requests	≥ 2.31	PubChem API requests	Apache 2.0
openpyxl	≥ 3.1	Excel read/write	MIT
Pillow	≥ 10.0	Icon (.ico) generation	PIL
re	stdlib	Regular expressions	PSF
json	stdlib	Settings and configuration	PSF
pathlib	stdlib	File path management	PSF
threading	stdlib	Non-blocking execution	PSF

7.2 Data sources

Source	URL	Data retrieved
MedChemExpress (MCE)	www.medchemexpress.com	Description, /Targets/ links, bibliographic references
PubChem PUG View	pubchem.ncbi.nlm.nih.gov/rest/pug_view	Full pharmacological data, references
PubChem simple API	pubchem.ncbi.nlm.nih.gov/rest/pug	Short description (fallback)
PubChem HTML	pubchem.ncbi.nlm.nih.gov/compound	Full HTML page (fallback)
canSAR.ai	cansar.ai/compound/{id}/target-affinity	Target proteins (Homo sapiens, Mean Potency < 1000 nM), via Playwright, simulated search navigation

7.3 Execution flow

• Reading the source file → extraction of (name, CID) pairs
• For each compound: opening the MCE page via Playwright (real Chromium browser)
• Extraction from MCE: /Targets/ links, description #product_syn, references #cpd_References1
• MCE fallback: if the direct URL fails, search by name and click on the first result
• PubChem query: PUG View → simple API → HTML
• Target merge: MCE first, PubChem as complement (no duplicates), columns "Target(s) MCE" and "Target(s) PubChem"
• If CanSAR is checked: search compound by name, read Target Affinity table filtered by Homo sapiens / Mean Potency < 1000 nM, written to the "Targets CanSAR" column (never merged with other target columns)
• Write to the Excel file with formatting
• Real-time update of the log and progress bar

8. User interface

8.1 Left panel (configuration)

Section	Content
1. Source file	CSV/Excel load button + CID column selector
2. Output folder	Folder selector for the generated Excel file
3. Keywords (optional)	Additional search terms + words to exclude

8.2 Right panel (execution and results)

Area	Role
Run button	Starts the search in a separate thread (UI remains responsive)
Progress bar	Displays % completion and current compound
Results table	Real-time preview of targets found per compound
Log	Detailed progress log, colour-coded by level (ok/warning/error)

8.3 Available languages

Flag	Language
🇫🇷	French (default)
🇬🇧	English
🇪🇸	Spanish
🇵🇹	Portuguese
🇩🇪	German
🇨🇳	Simplified Chinese

9. GDPR Compliance and Security

CTInfer was designed from the outset to comply with the requirements of the General Data Protection Regulation (GDPR, Regulation (EU) 2016/679) and the ANSSI recommendations for secure development.

CTInfer does not process any personal data. The data handled consists exclusively of public chemical identifiers (PubChem CIDs), compound names, and pharmacological information sourced from public scientific databases.

9.1 Nature of processed data

Data type	Nature	GDPR sensitivity
PubChem CID	Public numeric identifier	NONE: public data
Compound name	Chemical denomination	NONE: public data
Biological descriptions	Text extracted from public databases	NONE: public data
Bibliographic references	Public scientific citations	NONE: public data
Excel output file	Pharmacological research results	NONE: no personal data

9.2 GDPR principles applied

Data minimisation

CTInfer only collects information strictly necessary for identifying molecular targets: the compound name and its PubChem CID. No additional data is required or stored.

No transmission to unauthorised third parties

The only network flows from the application are to three public scientific databases: PubChem (NIH/NCBI, United States), MedChemExpress (commercial bioactive compound database) and, when the CanSAR source is enabled, canSAR.ai (MD Anderson Cancer Center). These queries are equivalent to a manual consultation from a web browser. No user data is transmitted.

No additional storage

CTInfer creates no local copies of data beyond the Excel results file generated in the folder chosen by the user. No local database, no cache, no persistent log is created.

No telemetry

The application contains no telemetry system, no external logging, and no connection to third-party cloud services. No usage statistics are collected or transmitted.

Traceability

The real-time log displayed in the interface constitutes a record of each operation performed during the session. This log is not saved automatically, it disappears when the application is closed.

9.3 Risk analysis

Identified risk	Level	Mitigation measure
Personal data leak	NONE	No personal data is processed by the application.
Unauthorised transmission	LOW	Only PubChem, MCE and, when enabled, canSAR.ai are queried, public/institutional databases. No other network flow.
Unauthorised access to results	LOW	The Excel output file is stored locally in the folder chosen by the user. Access depends on the workstation security policy.
Source unavailability	LOW	If PubChem or MCE is unavailable, corresponding cells remain empty. The application does not crash.
Playwright/Chromium dependency	LOW	Playwright is maintained by Microsoft (Apache 2.0). Chromium is Google's open source browser. Two world-class reference organisations.
Future MCE compatibility	MEDIUM	If MCE modifies its HTML structure, selectors may need updating. Mitigation: search fallback + regex patterns.
Future canSAR.ai compatibility	MEDIUM	If canSAR.ai modifies its HTML structure or search flow, Playwright selectors may need updating. Source can be disabled independently without affecting MCE/PubChem.
Uncontrolled update	LOW	Standalone application with no automatic updates. Any modification is manually validated.

9.4 Compliance with ANSSI recommendations

• Exclusive use of open source libraries maintained by recognised organisations (Python Software Foundation, Microsoft, NIH)

• No proprietary or opaque component in the processing chain

• Fully readable and auditable source code

• No privilege escalation required, the application runs in standard user mode

• No modification of the Windows registry or system services

• Dependencies installed only via pip (official PyPI channel)

9.5 Institutional adoption of technologies used

Technology	Publisher	Adoption references
Python	Python Software Foundation	Reference language in scientific research, bioinformatics, WHO
Playwright	Microsoft Corporation	Used in production by Microsoft, Google, Adobe, GitHub, LinkedIn
Chromium	Google Chromium Project	Open source base of Chrome, over 3 billion users
PubChem	NIH / NCBI (United States)	World reference public chemical database, >115 million compounds
MedChemExpress	MedChemExpress LLC	Referenced supplier of bioactive compounds for research
openpyxl	openpyxl project	De facto standard for Excel manipulation in Python, MIT license
canSAR.ai	MD Anderson Cancer Center (Texas)	Integrated cancer research knowledge base (biology, chemistry, pharmacology); academic reference for therapeutic target identification

10. References and sources

Source	URL and reference
Python Software Foundation	https://www.python.org/about/ Governance, license, security
Playwright Microsoft	https://playwright.dev Documentation, architecture, Apache 2.0 license
Chromium Security	https://www.chromium.org/Home/chromium-security/
PubChem NIH/NCBI	https://pubchem.ncbi.nlm.nih.gov Public chemical database
PubChem API PUG View	https://pubchem.ncbi.nlm.nih.gov/docs/pug-view
MedChemExpress	https://www.medchemexpress.com
GDPR EUR-Lex	https://eur-lex.europa.eu/eli/reg/2016/679 Regulation (EU) 2016/679
ANSSI Python Security Guide	https://www.ssi.gouv.fr Secure development recommendations
canSAR.ai MD Anderson	https://cansar.ai Cancer research knowledge base (targets, affinities, biology)

Document issued the 30 June 2026