Upload
vandana-kartik
View
17
Download
0
Tags:
Embed Size (px)
DESCRIPTION
Tenth International Workshop CIA 2006, Edinburgh. Evaluating Dynamic Services in Bioinformatics. Maíra R. Rodrigues Michael Luck University of Southampton, UK. Outline. Bioinformatics Agents and Bioinformatics Model for Cooperative Interactions: Overview - PowerPoint PPT Presentation
Citation preview
Evaluating Dynamic Services in Bioinformatics
Maíra R. Rodrigues
Michael Luck
University of Southampton, UK
Tenth International Workshop CIA 2006, Edinburgh
CIA 2006ECS - University of Southampton2
Outline
Bioinformatics Agents and Bioinformatics Model for Cooperative Interactions: Overview Requirements for Service Evaluation Evaluation Method Example Scenario Conclusion Future Work
CIA 2006ECS - University of Southampton3
Bioinformatics
Bioinformatics Application of computer technology to manage
and analyse biological data Bioinformatics Services
Heterogeneous Locally and remotely used Continuous update
Management and analysis of biological data and tools Suitability of an agent-based approach
CIA 2006ECS - University of Southampton4
Bioinformatics
Interrelated data Cooperative applications
Participants request and provide services to each other
Services free of charge Non-economic exchange of
different types of tools and data
Interactions are based on reciprocal relations
CIA 2006ECS - University of Southampton5
Agents and Bioinformatics
The agent-based approach: Agents provide and request bioinformatics
services Existence of alternative providers Services are provided with different levels of
quality (heterogeneity) Therefore..
Agents need to select service providers
CIA 2006ECS - University of Southampton6
Agents for Interaction
Agent-based applications in bioinformatics: Concerned with high-level management tasks
Our concern: Model non-economic cooperative interactions Evaluation method for bioinformatics services
to determine an agent’s satisfaction Guide agent’s decisions over service
providers
CIA 2006ECS - University of Southampton7
Model for Cooperative Interactions
A1 A2
service
Model non-economic cooperative interactions based on exchange values (Piaget 1973)
• satisfaction
• debt
• effort
• credit
CIA 2006ECS - University of Southampton8
Model for Cooperative Interactions
A1 A2• satisfaction
• debt
• effort
• credit
Model non-economic cooperative interactions based on exchange values (Piaget 1973)
A1 A2• credit
• satisfaction
• debt
• effort
CIA 2006ECS - University of Southampton9
Model for Cooperative Interactions Exchange values result from the agent’s
evaluation of the service
Partner Selection(future interactions)
Exchange ValuesService
Evaluation
CIA 2006ECS - University of Southampton10
Model for Cooperative Interactions Exchange values result from the agent’s
evaluation of the service Exchange values (Rodrigues, Luck 2005, 2006)
Current work focus on service evaluation
Partner Selection(future interactions)
Exchange ValuesService
Evaluation
CIA 2006ECS - University of Southampton11
Bio-Services are dynamic: Constant updates Regular behaviour, but Sensitive to different parameter configuration
Evaluation requires Repeated evaluation Attach context information Evaluation of different aspects of the service
Service Evaluation
CIA 2006ECS - University of Southampton12
Evaluation method should address: Generality: apply to different types of bio-
services and aspects of these services Continuity: repeat evaluation every time a
service is received Consistency: compare evaluations made at
different points in time Discriminated information: allow flexible
decision-making by using evaluation of individual aspects or a global evaluation
Service Evaluation
CIA 2006ECS - University of Southampton13
Alternative Approaches
Quantitative approaches Scoring or utility functions
• Objective values• Precision, consistency, combination is
straightforward Qualitative approaches:
Classification rules (e.g., poor, good, excellent)
• Subjective values
CIA 2006ECS - University of Southampton14
Evaluation Method
Choose evaluation attributes for service examples: performance, quality, reliability,
etc. For each attribute, associate result
measures Pieces of information derived from service
result that can determine the service utility in relation to an attribute (observed value).
Static or dynamic measures (e.g., quality of interface and response time)
CIA 2006ECS - University of Southampton15
Evaluation Method
General evaluation function for evaluation attributes (utility): For a set of attributes A = {a1,..,ai}
Ui = bc
result measure for ai
evaluation strictness
0
1
c
Ui
CIA 2006ECS - University of Southampton16
Evaluation Process
Before evaluation: Identify evaluation attributes for services and result
measures for each attribute
Repeat evaluation process every time a service is received Input is the service result and configuration used For each evaluation attribute ai
• Compute result measures
• Calculate evaluation Ui
• Store evaluation Output is a set of evaluations (evaluation tuple)
CIA 2006ECS - University of Southampton17
Evaluating Bio-Services
Proteomics research Protein identification services
• Input: file (list of unknown peptides)• Process: database + matching algorithm• Output: list of proteins, peptides per protein
Services: OMSSA, MASCOT, Tandem Local and Remote Heterogeneous results for same input data Sensitive to different input configurations Evaluation can be used as criterion for future
selection
CIA 2006ECS - University of Southampton18
Evaluation attributes: Sensitivity
• Capacity of matching related proteins Accuracy
• Capacity of identifying true matches Performance
• Time taken from input submission until result is received
Evaluating Bio-Services
CIA 2006ECS - University of Southampton19
Result measures (rm): Sensitivity
• Number of proteins• Peptide ratio - peptides per protein
• Influence of input size• Increasing utility
input_size
peptide_ratio x protein_number
Evaluating Bio-Services
rm =
CIA 2006ECS - University of Southampton20
Accuracy • Number of false positives
• Decreasing utility
rm = false_positives
Performance: • Response time
• Influence of input size• Decreasing utility
response_time
input_size
Evaluating Bio-Services
rm =
CIA 2006ECS - University of Southampton21
Evaluation functions:
Ui = 0.5rm
Sensitivity (U1):
• U1 increases with peptide_ratio and protein_number
Accuracy (U2):
• U2 decreases with false_positives
Performance (U3):
• U3 decreases with response_time
Evaluating Bio-Services
CIA 2006ECS - University of Southampton22
Evaluating Bio-Services
Practical evaluation: Same input spectra Two different configurations (C1 and C2)
Evaluation of sensitivity
Evaluation reflects different results for C1 and C2
CIA 2006ECS - University of Southampton23
Evaluation of performance
Again, evaluation reflects different results for C1 and C2
Evaluating Bio-Services
CIA 2006ECS - University of Southampton24
Conclusions
Present an evaluation method to be used by agents requesting dynamic services in bioinformatics
Discussion of issues for efficient evaluation of these services, including Adoption of a repeated evaluation process Absolute evaluations Generation of individual and compatible evaluations Single evaluation must be calculated during selection
CIA 2006ECS - University of Southampton25
Conclusions
Show the application of the evaluation method for protein identification services
Importance of dynamic (repeated) evaluation is shown through empirical results
Provide more accurate information for agents that need to select services with dynamic characteristics
CIA 2006ECS - University of Southampton26
Future Work
Develop selection strategies that use and combine service evaluations Combination through objective and
subjective values Probabilistic analysis of past evaluations Consider similarity between different service
configurations Validate evaluation results with those of
bioinformaticians
CIA 2006ECS - University of Southampton27
Thank you
CIA 2006ECS - University of Southampton28
References
J. Piaget. Sociological Studies. Routlege, London, 1973.
M. R. Rodrigues and M. Luck. Analysing partner selection through exchange values. In Jaime Sichman and Luis Antunes, editors, Multi-Agent-Based Simulation VI, volume 3891 of Lecture Notes in Artificial Intelligence, pages 24-40, Berlin Heidelberg, 2006a. Springer-Verlag.
M. R. Rodrigues and M. Luck. Cooperative interactions: An exchange values model. In Coordination, Organization, Institutions and Norms in Agent Systems (COIN), ECAI Conference, Riva del Garda, Italy, August 2006b.