BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Machine Learning Models for Electron-Impact Ionization Cross Secti
 ons and Mass Spectra
DTSTART;VALUE=DATE-TIME:20260908T081000Z
DTEND;VALUE=DATE-TIME:20260908T084000Z
DTSTAMP;VALUE=DATE-TIME:20260922T191354Z
UID:indico-contribution-2040@events.ncbj.gov.pl
DESCRIPTION:Speakers: Bartosz Majewski (Gdańsk University of Technology)\
 nElectron-impact processes are fundamental to analytical chemistry\, plasm
 a physics\, radiation science\, and astrochemistry\, yet experimental meas
 urements remain costly and incomplete. We present machine learning approac
 hes for predicting two key electron-impact observables directly from molec
 ular structure: electron ionization mass spectra (EI-MS) and total electro
 n-impact ionization cross-sections.\n\nFor EI-MS prediction\, we developed
  a hybrid model combining a graph neural network encoder\, a residual neur
 al network decoder\, cross-attention refinement\, bidirectional prediction
 \, and chemistry-informed masking. Trained on the NIST14 EI-MS database (m
 olecules ≤500 Da)\, the model achieves strong library-matching performan
 ce (Recall@10 ≈ 80.8%). For ionization cross-sections\, we introduce a s
 tacked ensemble integrating LightGBM\, CatBoost\, a multilayer perceptron\
 , and a graph neural network\, combined through a residual LightGBM meta-m
 odel. Trained on 212 molecules from the Astrochemistry Low-energy Electron
  Cross-Section database\, the ensemble achieves a test-set RMSE of 1.160 a
 ₀² and a Spearman correlation of 0.990\, outperforming all individual m
 odels.\n\nThe results demonstrate that graph-based molecular representatio
 ns enable accurate prediction of complementary electron-impact properties.
  These approaches can augment reference databases\, support molecular iden
 tification\, and reduce the need for costly experimental measurements.\n\n
 https://events.ncbj.gov.pl/event/468/contributions/2040/
LOCATION:
URL:https://events.ncbj.gov.pl/event/468/contributions/2040/
END:VEVENT
END:VCALENDAR