1 of 81

CIFARP2025 – October, 2025

Fernando Luís Barroso da Silva, PhD

Department of Biomolecular Sciences

School of Pharmaceutical Sciences at Ribeirão Preto – USP

flbarroso@usp.br

Artificial Intelligence and Physics-based Simulation in Antibody Engineering as a Bridge between Research and Education

Fernando Barroso, USP

October, 2025

2 of 81

Goal for today

Institutional models and future reflections

Identifying the pedagogical gap

Why AI education is an immediate necessity for pharmaceutical sciences

From Antibody Engineering research to classroom strategies

Fernando Barroso, USP

October, 2025

3 of 81

The AI Imperative (I)

From Future Concept to Present Reality

3

flbarroso@usp.br

4 of 81

The AI Imperative (I)

From Future Concept to Present Reality

https://arstechnica.com/information-technology/2023/06/chatgpt-takes-the-pulpit-ai-leads-experimental-church-service-in-germany/

https://www.youtube.com/watch?v=8P9oSgrT35o

Societal Integration

4

flbarroso@usp.br

5 of 81

The AI Imperative (I)

AI Agents are moving from conceptual models to practical applications

https://ceas.uc.edu/research/centers-labs/

center-for-smart-sustainable-and-resilient-infrastructure/research.html

5

flbarroso@usp.br

6 of 81

The AI Imperative (II)

Revolutionizing the Research Pipeline?

The AI Scientist

6

flbarroso@usp.br

7 of 81

The AI Imperative (II)

Revolutionizing the BioPharma Pipeline

https://www.nvidia.com/en-us/customer-stories/astellas-antibody-language-model-with-bionemo/

7

flbarroso@usp.br

8 of 81

The AI Imperative (II)

Revolutionizing the BioPharma Pipeline

8

flbarroso@usp.br

9 of 81

The AI Imperative (II)

How far are we?

9

flbarroso@usp.br

10 of 81

The AI Imperative (III)

The Industry Demand for "AI-Ready" Graduates

Artificial intelligence (AI) offers potential benefits for the design and operation of bioprocesses. This workshop will bring 150 participants from academia and industry to Los Angeles on Sept. 26-27 (2024). They will develop recommendations for bioscience-related AI education and training strategies.

https://www.nsf.gov/awardsearch/showAward?AWD_ID=2432457&HistoricalAwards=false

A critical skills gap has been identified by industry leaders

10

flbarroso@usp.br

11 of 81

The AI Imperative (III)

The Industry Demand for "AI-Ready" Graduates

The biopharmaceutical industry requires a workforce trained in AI literacy now.

11

flbarroso@usp.br

12 of 81

The Pedagogical Challenge

Diagnosing the "AI-Native" Student

Observation 1: AI as an Oracle, Not a Tool

Students demonstrate rapid adoption of generative AI, but often treat it as an infallible source of truth rather than a co-pilot.

Evidence: Replacing foundational learning (e.g., textbook chapters) with LLM-generated summaries, which may contain factual "hallucinations“.

Evidence: Chats + Submission of correct solutions to problem sets (e.g., in Physical Chemistry) without the ability to demonstrate the underlying fundamental reasoning.

12

flbarroso@usp.br

13 of 81

The Pedagogical Challenge

Diagnosing the "AI-Native" Student

Observation 2: Homogenization of Critical Texts/Analysis

Evidence: A marked increase in the similarity of analytical sections in laboratory reports and essays.

Warning: This suggests a convergence of student output driven by reliance on similar prompts, leading to a potential atrophy of independent critical thought and synthesis.

13

flbarroso@usp.br

14 of 81

The Pedagogical Challenge

Diagnosing the "AI-Native" Student

The challenge is not one of academic integrity alone.

It is a significant pedagogical gap: Students possess powerful tools without the critical framework (e.g., training, prompt engineering, validation, ethical context, etc) required for their effective and responsible use in healthcare sciences.

Hypothesis

Naya Abdallah, Rateb Katmah, Kinda Khalaf, Herbert F. Jelinek,

Systematic review of ChatGPT in higher education: Navigating impact on learning, wellbeing, and collaboration, Social Sciences & Humanities Open, Volume 12, 2025, 101866

14

flbarroso@usp.br

15 of 81

Practical strategies

The Intervention

Fernando Barroso, USP

October, 2025

16 of 81

The Intervention (I)

Building Critical Awareness

Semantic Ambiguity in Multimodal Prompting

Gere uma imagem do Batman plantando bananeira

16

flbarroso@usp.br

17 of 81

The Intervention (I)

Building Critical Awareness

Semantic Ambiguity in Multimodal Prompting

Gere uma imagem do Batman plantando bananeira

The responsibility for the output's validity rests entirely with the user.

17

flbarroso@usp.br

18 of 81

The Intervention (I)

Building Critical Awareness

Semantic Ambiguity in Multimodal Prompting

A photograph of a boxer hitting the sack after a long day of training

18

flbarroso@usp.br

19 of 81

https://www.raps.org/news-and-articles/news-articles/2025/10/fda-officials-high-quality-data-is-essential-for-a

19

flbarroso@usp.br

20 of 81

Promises and Pitfalls in Bioinformatics

The Intervention (II)

Fernando Barroso, USP

October, 2025

21 of 81

Structural Biology:

A Field Redefined!!

Fernando Barroso, USP

October, 2025

22 of 81

And Neither Has Medicine!

Fernando Barroso, USP

October, 2025

23 of 81

Quality of AlphaFold structures?

(Nejedlá, Košovan & Barroso, in preparation)

Simulation data

Fernando Barroso, USP

October, 2025

24 of 81

Antigen-antibody interfaces

(Barroso da Silva, Paco, Laaksonen & Ray, Biophys. R, 2025)

Simulation data

Fernando Barroso, USP

October, 2025

25 of 81

1OAN

pH 8.5

1OK8

pH 7.0

Structure of the Dengue 2 virus envelope protein

prefusion

post fusion

Deposited: 2003-01-16 Released: 2003-05-16

Deposited: 2003-07-19 Released: 2004-01-29

> ENVELOPE GLYCOPROTEIN|DENGUE VIRUS TYPE 2 (11060)MRCIGISNRDFVEGVSGGSWVDIVLEHGSCVTTMAKNKPTLDFELIKTEAKQPATLRKYCIEAKLTNTTTESRCPTQGEPTLNEEQDKRFVCKHSMVDRGWGNGCGLFGKGGIVTCAMFTCKKNMEGKIVQPENLEYTVVITPHSGEEHAVGNDTGKHGKEVKITPQSSITEAELTGYGTVTMECSPRTGLDFNEMVLLQMKDKAWLVHRQWFLDLPLPWLPGADTQGSNWIQKETLVTFKNPHAKKQDVVVLGSQEGAMHTALTGATEIQMSSGNLLFTGHLKCRLRMDKLQLKGMSYSMCTGKFKVVKEIAETQHGTIVIRVQYEGDGSPCKIPFEIMDLEKRHVLGRLITVNPIVTEKDSPVNIEAEPPFGDSYIIIGVEPGQLKLNWFKK

Fernando Barroso, USP

October, 2025

26 of 81

1OAN

pH 8.5

prefusion

1OK8

pH 7.0

post fusion

Structure of the Dengue 2 virus envelope protein

pTM = 0.76

100%

0%

Where Is the pH Dependence???

[Barroso da Silva et al, JCIM, 2024]

"The native conformation is determined by the totality of interatomic interactions and hence by the amino acid sequence, in a given environment.“ (Christian B. Anfinsen)

Fernando Barroso, USP

October, 2025

27 of 81

Dengue 2 virus

envelope protein

[FLBDS & CE, in preparation]

1OAN

pH 8.5

1OK8

pH 7.0

prefusion

post fusion

Why simulation Methods still Matter in Biomolecular Research

Physical-based simulation

Fernando Barroso, USP

October, 2025

28 of 81

AI alone is insufficient and must be integrated with physics-based simulation

But….

Fernando Barroso, USP

October, 2025

29 of 81

Physical-based simulation

Towards better mAbs (Pipeline 1)

Electrostatic optimization

Fernando Barroso, USP

October, 2025

30 of 81

Affinity?

Epitopes?

Aggregation?

Stability?

Theoretical ALA scanning

FGEVFN…

AGEVFN…

FAEVFN…

FGAVFN…

FGEAFN…

FGEVAN…

FGEVFA…

Towards better mAbs

Computer tools to short development time

for SARS-CoV-2 treatment and prevention

data volume

mAb

l

r

Electrostatic optimization

Fernando Barroso, USP

October, 2025

31 of 81

Free energy profile for the interaction of SARS-CoV-2 S RBD proteins with a new (optimized) monoclonal antibody

[Giron, Laaksonen & FLBDS, Virus Research, 2020]

CR3022’

binding affinity

SARS-CoV1 SARS-CoV2

CR3022 27 33

Number of aa affected

CR3022´ -- 34

Simulation data

Vaccines 2021, 9(12), 1409

Fernando Barroso, USP

October, 2025

32 of 81

Physical-based simulation

Towards better mAbs (Pipeline 2)

Multiple scales in silico protocol

Fernando Barroso, USP

October, 2025

33 of 81

Scheme for the multiple scales in silico protocol for antibody design (III)

CG MD

SIRAH

AA MD

TAS

RBD (variants)

Evaluate Rosetta Score Function using MD

(free energy analysis)

Evaluate Rosetta Score Function using MC

(free energy analysis)

Consensus

To enhance our understanding of the close interactions of the mAb-RBDwt

Electrostatic optimization

Binding affinities with other RBD

[Neamtu, Mocci, Laaksonen & Barroso da Silva, Colloids Surf. B: Biointerfaces 2024]

Accelerating Antibody Engineering Through Computer Tools

Fernando Barroso, USP

October, 2025

34 of 81

Design new binders:

binding affinity

candidates

Constant-charge MD

(SIRAH+umbrella sampling)

[Neamtu, Mocci, Laaksonen & Barroso da Silva, Colloids Surf. B: Biointerfaces 2024]

Testing top-10 mAbs designed by RAbD

Simulation data

Fernando Barroso, USP

October, 2025

35 of 81

Machine learning approach

….It allows you to go further!

Fernando Barroso, USP

October, 2025

36 of 81

Design new binders

[Paco et al, in preparation - https://openreview.net/pdf?id=lpMBF5kNVX]

Antibody Specificity Predictor (ASPred)

flbarroso@usp.br

37 of 81

Design new binders

[Paco et al, in preparation - https://openreview.net/pdf?id=lpMBF5kNVX]

Antibody Specificity Predictor (ASPred)

These AI-generated candidates are predicted to possess superior binding affinity compared to those designed using only physics-based or traditional directed evolution approaches.

They generate plausible sequences but require biophysical validation.

flbarroso@usp.br

38 of 81

Experimental data

Simulation data

[Paco et al, in preparation]

Simulation data

flbarroso@usp.br

39 of 81

The Intervention (III)

Translating the "AI + Validation" framework into classroom activities

Course: Undergraduate Physical Chemistry (Thermodynamics)

Objective: To develop and apply critical thinking skills to evaluate AI-generated solutions.

Activity: "The AI Fact-Check" (Problem Set)

Step 1 (Human Solution): Students must first solve a problem set (e.g., Laws of Thermodynamics) on their own.
Step 2 (AI Solution): Students then prompt a unique LLM to solve the entire problem set.
Step 3 (Critical Analysis): The core task is to identify and explain at least one significant conceptual or mathematical error in the AI's output for a chosen exercise.
Deliverable: Students submit their corrected solution, a detailed error analysis, and the full, unedited AI interaction transcript.

Outcome: This process forces students to move from "passive acceptance" to "active critical validation”, reinforcing that the human expert is the final arbiter of truth.

39

flbarroso@usp.br

40 of 81

The Intervention (III)

Translating the "AI + Validation" framework into classroom activities

Course: Graduate Topics in Health Bioinformatics

Objectives:

a) Mastering the physical principles of computational validation.

b) Effectively utilizing LLMs as a "research co-pilot".

Activity: "MD Simulation: Protocol Rationale & Adaptation" (Problem Set)

Task: Students execute an MD protocol, focusing on the underlying physical rationale of each step.
AI Instruction: LLMs are encouraged for brainstorming conceptual points and troubleshooting, but students must critically evaluate and verify all information.

Outcome: Trains students for R&D roles by developing both mastery of physics-based validation and the critical, effective use of AI as a research partner.

40

flbarroso@usp.br

41 of 81

The Intervention (IV)

Post-Intervention Survey Profile

Question 1: Frequency of Use

How frequently do you utilize generative AI tools (e.g., ChatGPT, Gemini) to assist with your university coursework and studies?

Daily
Several times per week
Several times per month
Rarely (only for very specific tasks)
Never

Key Finding 1: High Adoption, Shaped Usage

Student engagement with AI remains high:

80% of users report daily or weekly frequency.

Our pedagogical goal is not to prevent usage, but to shape it.

41

flbarroso@usp.br

42 of 81

The Intervention (IV)

Post-Intervention Survey Profile

Question 2: Application (Purpose of Use)

List the top three (3) primary academic purposes for which you use these AI tools. Be brief and objective. (Examples: clarifying concepts, summarizing articles, translating texts, rewriting paragraphs, generating ideas for assignments, etc.)

Key Finding 2: Applications Align with "Tutor" Perception

Primary uses are cognitive and support-based, not simple output generation:

Conceptual Clarification: 81.8%
Exercise Resolution: 45.5% (!)

42

flbarroso@usp.br

43 of 81

The Intervention (IV)

Post-Intervention Survey Profile

Question 3: Workflow Process (Critical Thinking)

Imagine you are using an AI to develop a section of a laboratory report on a concept you do not fully understand. Which of the options below best describes your most common process?

I use the text generated by the AI directly in my work, making only minor edits to style and format.
I read the text generated by the AI to understand the subject and then rewrite it in my own words, trusting the accuracy of the information presented by the tool.
I use the AI text as a foundation, but I verify the most important information and data against reliable sources (textbooks, articles, course materials) before finalizing my writing.
I primarily use the AI to get an initial summary, generate ideas, or structure my thoughts, but I develop the content of my report primarily from the course-provided sources.

43

flbarroso@usp.br

44 of 81

The Intervention (IV)

Post-Intervention Survey Profile

Question 3: Workflow Process (Critical Thinking)

Imagine you are using an AI to develop a section of a laboratory report on a concept you do not fully understand. Which of the options below best describes your most common process?

Key Finding 3: A Dramatic Shift in Process?

Students report a high degree of critical validation in their workflow:

72.7%: Use AI as a foundation, but verify with reliable sources (textbooks, articles).
18.2%: Use primarily for ideation, relying on course sources for content.
0%: Reported using AI-generated text directly ("copy/paste").

44

flbarroso@usp.br

45 of 81

The Intervention (IV)

Post-Intervention Survey Profile

Question 4: Perception of Role

Which of the following options best describes how you perceive the primary role of generative AI in your studies?

a) A source of direct and ready answers, similar to an advanced search engine or an encyclopedia. b) A writing assistant, focused on improving the grammar, style, and clarity of my texts.

c) A personalized tutor, capable of explaining complex concepts in different ways to facilitate my learning.

d) A tool for brainstorming and ideation, which helps me explore different approaches to a problem or assignment.

Key Finding 4: A Consensus Shift in Perception

There is a clear consensus on the tool's role following the intervention.

Personalized Tutor: 90.9%
Only 9.1% view it as a simple "source of direct answers" (an encyclopedia).

45

flbarroso@usp.br

46 of 81

The vision

Fernando Barroso, USP

October, 2025

47 of 81

Education 1

“Transforming the future of engineering research and

education through Applied AI.”

[https://engr.ncsu.edu/applied-ai/]

Organizing for AI

flbarros@ncsu.edu

Fernando Barroso, USP & NCSU

September, 2024

48 of 81

Education 2

ICTP-SAIFR Advanced School (São Paulo, Brazil)

______________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

April 14-18, 2025

School on Biological Physics and Biomolecular Simulations in the Machine Learning Era

Fernando Luís Barroso da Silva

NCSU & USP

Ralf Eichhorn

NORDITA & SU

https://www.ictp-saifr.org/2025-activities/

flbarros@ncsu.edu

Fernando Barroso, USP & NCSU

September, 2024

49 of 81

Education 3

https://moritzlaw.osu.edu/CRAIG

flbarros@ncsu.edu

Fernando Barroso, USP & NCSU

September, 2024

50 of 81

Some Reflections

From AI-User to AI-Ready Professional

Fernando Barroso, USP

October, 2025

51 of 81

Reflections

The goal is not simply to teach students about AI, but to teach them how to

think effectively and ethically in a world with AI.

Foundational Knowledge is the Validator

The Co-Pilot Needs a Pilot

The "Brain Academy"

A Co-pilot is useless (and dangerous!) without a trained, certified, and responsible human Pilot

Our role is to train the Pilot (including how to interact with the Co-Pilot)!

As AI automates routine cognitive tasks, our value shifts to higher-order critical thinking

51

flbarroso@usp.br

52 of 81

Reflections

52

flbarroso@usp.br

53 of 81

I realized that my role had shifted from scripting to supervising. What matters now is stating the question clearly, spotting problems that the computer cannot see, and taking responsibility for the answer.

53

flbarroso@usp.br

54 of 81

https://www.facebook.com/labbpc

Join us!

Carolina Correa Giron (UFTM/Brazil)
Sergio Alejandro Poveda (FCFRP/Brazil)
Ilyas Grandguillaume (FCFRP/Brazil & UPC/France)

Dr. Animesh Ray and his team (KGI&Caltech/USA)

Dr. Catherine Etchebest and her team (USPC/France)

Dr. Erik Santiso and his team (NCSU/USA)

Dr. Aatto Laaksonen and his team (SU/Sweden)

Dr. Peter Kosovan and his team (CUNI/Czech)

Many thanks!

Thank you for your time!

Fernando Barroso, USP

October, 2025

55 of 81

Combining physical-based simulation methods and machine learning approaches for Advanced Antibody Engineering

RAGSAB – September 26, 2024

Fernando Luís Barroso da Silva, PhD

Department of Biomolecular Sciences – USP/Brazil

Department of Chemical and Biomolecular Engineering – NCSU/US

flbarros@ncsu.edu

Fernando Barroso, FCFRP/USP

October, 2025

56 of 81

models

&

computer tools

biomolecular interactions

virus & binders

diagnosis

treatment

prevention

The molecular basis for understanding diseases and designing pharmaceuticals, bioseparation processes, and new functionalized (bio)materials...

Development of tools
Understanding

molecular mechanisms

Guide innovations

Lunkad, Kosovan & Barroso, JACS 144 (4), 1813-1825

Fernando Barroso, FCFRP/USP

October, 2025

57 of 81

General approach: to rationalize key applied systems

pH and salt

Example of measurements

microencapsulation

“toy model”

+

-

+

-

[Jonsson, Lund & Barroso da Silva, Food Colloids: Self-Assembly and Material Science, 2007; Poveda, Etchebest & Barroso da Silva, J. Chem. Inf. Model, 2020]

57

flbarroso@usp.br

58 of 81

General approach: to rationalize key applied systems

[Jonsson, Lund & Barroso da Silva, Food Colloids: Self-Assembly and Material Science, 2007; Poveda, Etchebest & Barroso da Silva, J. Chem. Inf. Model, 2020]

pH and salt

Example of measurements

Complex biological system

“toy model”

Epitopes

It needs to be faster!

Binding affinities

pH-responsive antibodies

58

flbarroso@usp.br

59 of 81

structure 1

structure n

...

New

data

bank

Computational

Chemistry

Fernando Barroso, USP & NCSU

September, 2024

60 of 81

How to reduce cpu time in order to simulate large protein aggregates?�

Fernando Barroso, USP & NCSU

September, 2024

61 of 81

6-phosphogluconate

dehydrogenase

(PDB id 2zyg)

9185.2s

AMD Opteron 2356

processador (8 cores and 2.3 GHz)

Intel i7-3630QM and 2.40 GHz

96s

Performance PB FPTS

FPTS

[FLBDS & DMK, JCTC, 2017]

Fernando Barroso, USP & NCSU

September, 2024

62 of 81

Proteins

FPTS

[FLBDS & DMK, JCTC, 2017]

A machine learning model for titration (trained with 612 non-redundant cases) can achieve similar performance

(Barroso & Santiso, in preparation)

Fernando Barroso, USP & NCSU

September, 2024

63 of 81

Quality of AlphaFold structures?

(Nejedlá, Košovan & Barroso, in preparation)

63

flbarroso@usp.br

64 of 81

Antigen-Antibody interface

PPI in general

> No

ionizable aa

Viruses

&

Abs

Excelent system to study PPI!

and electrostatic interactions!

64

flbarroso@usp.br

65 of 81

Antibody-Antigen Interactions:

Insights from an analysis of epitope specificity and interface characteristics

For PPI:

filter

1363 complexes

J. Mol. Biol. (2010) 403, 660–670

(Grandguillaume, Etchebest & Barroso, in preparation)

Charged amino acids (ARG, GLU, ASP, and LYS) are up to 1.5 times more represented in epitopes than in PPI

Fernando Barroso, USP & NCSU

September, 2024

66 of 81

Antibody-Antigen Interactions:

Insights from an analysis of epitope specificity and interface characteristics

Using residue frequency as a descriptor of the interface and a random forest model, we examined the capacity to differentiate epitopes from non-immune protein-protein interfaces.

(Grandguillaume, Etchebest & Barroso, in preparation)

Scoring with learned models (1 & 2)

Can we find near native?

Fernando Barroso, USP & NCSU

September, 2024

67 of 81

Compass to the Severity of the Future Variants with the Charge-Rule

The slope of the linear regression is numerically in the same order as DVLO analytical approach:

[Barroso da Silva, Giron & Laaksonen, J. Phys Chem B, 2022]

binding affinity

67

flbarroso@usp.br

68 of 81

Design new binders

[Paco et al, in preparation - https://openreview.net/pdf?id=lpMBF5kNVX]

Antibody Specificity Predictor (ASPred)

flbarroso@usp.br

69 of 81

Design new binders

[Paco et al, in preparation - https://openreview.net/pdf?id=lpMBF5kNVX]

Antibody Specificity Predictor (ASPred)

flbarroso@usp.br

70 of 81

Design new binders

Antibody Specificity Predictor (ASPred)

[Paco et al, in preparation - https://openreview.net/pdf?id=lpMBF5kNVX]

Simulation data

flbarroso@usp.br

71 of 81

Experimental data

Simulation data

[Paco et al, in preparation]

Simulation data

flbarroso@usp.br

72 of 81

Design new binders

Antibody Specificity Predictor (ASPred)

[Paco et al, in preparation - https://openreview.net/pdf?id=lpMBF5kNVX]

Simulation data

flbarroso@usp.br

73 of 81

Design new binders

binding affinity

candidates

Constant-charge MD

(SIRAH+umbrella sampling)

[Neamtu, Mocci, Laaksonen & Barroso da Silva, Colloids Surf. B: Biointerfaces 2024]

Fernando Barroso, USP & NCSU

September, 2024

74 of 81

Design new binders

RAbD

After additional

electrostatic optimization

[Neamtu, Mocci, Laaksonen & Barroso da Silva, Colloids Surf. B: Biointerfaces 2024]

Fernando Barroso, USP & NCSU

September, 2024

75 of 81

Education 1

“Transforming the future of engineering research and

education through Applied AI.”

[https://engr.ncsu.edu/applied-ai/]

Fernando Barroso, USP & NCSU

September, 2024

76 of 81

Education 2

Tools and concepts employed in pharmaceutical and biotechnological industrial processes (undergraduate)

ICTP-SAIFR Advanced School (São Paulo, Brazil)

______________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

April 14-18, 2025

School on Biological Physics and Biomolecular Simulations in the Machine Learning Era

Fernando Luís Barroso da Silva

NCSU & USP

Ralf Eichhorn

NORDITA & SU

https://www.ictp-saifr.org/2025-activities/

Fernando Barroso, USP

October, 2025

77 of 81

Simplified models can aid the understanding of complex molecular

mechanisms. They can also be used to design new rational applications.

SARS-CoV-2 is undergoing an evolutionary process that enhances its

electrostatic properties.

Fast constant-pH simulation methods represent a promising frontier in in

addressing global health challenges and antibody engineering. They

can be adapted to a wide range of diseases and pathogens.

A highly efficient and robust multiple-scale in silico protocol can be achieved by integrating a variety of computational tools.

Highlights

flbarroso@usp.br

78 of 81

Thank you for your time!

Fernando Barroso, USP

October, 2025

79 of 81

The candidate must have an Ab-Ag interface score

below –150 REU (the native interface score is -65 REU).

The candidate must have an Ab-Ag interface surface

area larger than 1900 Å²

(the native complex interface surface area is 2060 Å²).

[Neamtu, Mocci, Laaksonen & Barroso da Silva, Colloids Surf. B: Biointerfaces 2024]

Scheme for the multiple scales in silico protocol for antibody design (I)

Fernando Barroso, USP

October, 2025

80 of 81

The candidate must have an Ab-Ag interface score

below –150 REU (the native interface score is -65 REU).

The candidate must have an Ab-Ag interface surface

area larger than 1900 Å²

(the native complex interface surface area is 2060 Å²).

native false positive

[Neamtu, Mocci, Laaksonen & Barroso da Silva, Colloids Surf. B: Biointerfaces 2024]

Scheme for the multiple scales in silico protocol for antibody design (II)

Fernando Barroso, USP

October, 2025

81 of 81

Design new binders

RAbD

After additional

electrostatic optimization

[Neamtu, Mocci, Laaksonen & Barroso da Silva, Colloids Surf. B: Biointerfaces 2024]

Still work to do!

Simulation data

Fernando Barroso, USP

October, 2025