Thermodynamic profiling of proteinligand binding energies - - PowerPoint PPT Presentation

▶
thermodynamic profiling of protein ligand binding energies
SMART_READER_LITE
LIVE PREVIEW

Thermodynamic profiling of proteinligand binding energies - - PowerPoint PPT Presentation

Mar 25, 2023 316 likes •572 views

Thermodynamic profiling of proteinligand binding energies Application of machine learning methods in Bioinformatics

slide-1
SLIDE 1

Thermodynamic profiling of proteinligand binding energies

Application of machine learning methods in Bioinformatics

slide-2
SLIDE 2
  • Background
  • Challenges
  • Tools / methods
  • Results / insights
  • Summary
slide-3
SLIDE 3
  • Human body is constantly invaded by

pathogens.

  • ”Proteins” on the surface of pathogens are

vital in adhesion and proliferation.

slide-4
SLIDE 4
  • !"

# $ %$ &''((''

slide-5
SLIDE 5
  • Drugs / Inhibitors / ligands

– Small molecules that prevent the adhesion or proliferation of pathogens.

  • ”Relenza” is the trade name for infuenza virus

inhibitors (ligand) that binds to a surface protein of influenza virus that stops proliferation.

slide-6
SLIDE 6
  • Host

Receptor Inhibitors Two important properties of a drug

  • Affinity
  • Specificity

(

slide-7
SLIDE 7
  • Affinity

– How strong does a drug bind to the target.

Specificity

– How specific are the drug’s interactions to that particular target

  • Is it binding to other proteins in the human body ?
  • Main cause of sideeffects
slide-8
SLIDE 8
  • Binding energy

– Strength of interaction between the protein and ligand (negative value indicates binding) ∆G = ∆ H T ∆ S

Enthalpy term Entropy term Binding energy

slide-9
SLIDE 9
  • %∆G

)'*(

  • (

%∆G )'*(

slide-10
SLIDE 10
  • ∆G = 10 kcal/mol

∆G = ∆ H T ∆ S Accurate estimation of ∆ H and T ∆ S is necessary for precise placement of ∆ G case

slide-11
SLIDE 11

+

Qualitative classification

– Neural networks – Support vector machines

Quantitative estimation

– Support vector machine regression

slide-12
SLIDE 12

+

Neural networks (Multilayered perceptron)

Each node/neuron in hidden layer is a nonlinear activation function

  • −

+ =

  • Where,

yi is output of neuron i, Si is the weighted sum of all inputs and bias to neuron i Error backpropagation algorithms

slide-13
SLIDE 13

+

Support vector machines classification

“Hyperplane” at the largest distance between border samples (“support vectors”) Hyperplane Support vectors

slide-14
SLIDE 14

+

Support vector machines classification

slide-15
SLIDE 15

+

SVM regression

slide-16
SLIDE 16

),

)'( (

  • +"./-
  • "!//-
  • ../)'

*0(

  • (
slide-17
SLIDE 17

)%

Feature selection / elimination

  • Feature reduction using Principal Component Analysis (PCA)

– Reduce the dimensionality of the data by fewer samples but still preserving the variance

  • Backward feature elimination (BFE)

Cross validation

– 2 fold Cross validation – Leaveoneout Cross validation – Nfold Stratified sampling cross validation

slide-18
SLIDE 18

Results Classification

slide-19
SLIDE 19

Confusion matrix

%&1

slide-20
SLIDE 20

Results Classification

slide-21
SLIDE 21

Results – SVR prediction models

+

+

  • Kernel and Hyper parameters
  • Cross validation methods
  • Outlier analysis
slide-22
SLIDE 22

Results – SVR prediction models

  • Outlier analysis
  • Correlation coefficient and
  • Standard error
slide-23
SLIDE 23

Application of SVR to real data

  • Kernel choice and parameters
  • Linear
  • Polynomial (degree)
  • Gaussian (width parameter)
  • Hyperparameters
  • Parameter C
  • Parameter ε

2#(-

  • Data Normalization (?)
slide-24
SLIDE 24

Summary

+"./-3../)'

  • "!//-

MLPNN and SVM Classifiers SVM regression models

slide-25
SLIDE 25

Thanks for listening !