Experiment Orchestration with the `optimize` Block¶

Introduction¶

The optimize block represents one of the most significant advances in GeneForgeLang, transforming the language from a simple specification tool to a complete scientific discovery orchestration platform. This block automates the exploration of parameter spaces to find optimal experimental conditions, inspired by 'AI-driven experimentation' techniques that are revolutionizing genomic research.

Instead of running experiments sequentially with fixed parameters, the optimize block allows defining intelligent experimental loops that:

Learn from each executed experiment
Adapt parameter selection based on previous results
Converge toward optimal experimental conditions efficiently
Minimize the number of experiments needed to find solutions

This approach is especially valuable in genomics, where experiments can be costly, slow, and where the parameter space is typically very large and complex.

Block Structure¶

The optimize block consists of five main components that fully define an experimental optimization loop:

`search_space` - Search Space¶

Defines the parameters to be explored and their possible value ranges. Each parameter can be:

Continuous: Using the syntax range(min, max) for numerical values
Discrete: Using the syntax choice([value1, value2, ...]) for specific options

Syntax:

search_space:
  parameter_name: range(min_value, max_value)
  another_parameter: choice([option1, option2, option3])

Examples:

search_space:
  temperature: range(25, 42)              # Temperature in °C
  guide_concentration: range(10, 100)      # Concentration in nM
  incubation_time: choice([6, 12, 24, 48])  # Hours of incubation
  buffer_type: choice([PBS, HEPES, Tris])     # Buffer types

`strategy` - Optimization Strategy¶

Specifies the artificial intelligence algorithm that will be used to guide the exploration of the parameter space. The name field is mandatory and identifies the optimization plugin to use.

Supported Algorithms:

ActiveLearning: Selects experiments that maximize information gained
BayesianOptimization: Uses Gaussian processes to model the parameter space
GeneticAlgorithm: Evolves populations of experimental configurations
SimulatedAnnealing: Stochastic exploration with gradual cooling
RandomSearch: Random selection (baseline)
GridSearch: Systematic exhaustive exploration

Syntax:

strategy:
  name: AlgorithmName
  specific_parameter: value
  another_parameter: value

Examples:

# Active learning with uncertainty metric
strategy:
  name: ActiveLearning
  uncertainty_metric: entropy
  initial_samples: 5

# Bayesian optimization with specific acquisition function
strategy:
  name: BayesianOptimization
  acquisition_function: expected_improvement
  kernel: rbf

`objective` - Optimization Objective¶

Defines which metric to optimize. It must contain exactly one of the following keys:

maximize: To maximize a metric (e.g.: efficiency, performance)
minimize: To minimize a metric (e.g.: cost, time, toxicity)

Optionally, it can include a target field that specifies the objective context.

Syntax:

objective:
  maximize: metric_name
  # Or alternatively:
  minimize: metric_name
  target: optional_context

Examples:

# Maximize editing efficiency
objective:
  maximize: editing_efficiency

# Minimize off-target effects
objective:
  minimize: off_target_effects
  target: complete_genome

# Maximize gene expression
objective:
  maximize: expression_level
  target: GFP_protein

`budget` - Budget and Stopping Criteria¶

Establishes constraints and stopping criteria for the optimization loop. It must contain at least one constraint.

Available Constraints:

max_experiments: Maximum number of experiments (integer)
max_time: Maximum time (format: "24h", "7d", "30m")
max_cost: Maximum budget (number)
convergence_threshold: Convergence threshold (0.0 - 1.0)

Syntax:

budget:
  max_experiments: number
  max_time: "time_with_unit"
  max_cost: amount
  convergence_threshold: threshold

Examples:

# Multiple limits
budget:
  max_experiments: 100
  max_time: 72h
  max_cost: 15000
  convergence_threshold: 0.01

# Only experiment limit
budget:
  max_experiments: 50

`run` - Execution Block¶

Defines the experiment or analysis that will be executed in each iteration of the optimization loop. It must contain exactly one of:

experiment: A standard GFL experiment block
analyze: A standard GFL analysis block

Parameters defined in search_space can be injected using the syntax ${parameter_name}.

Syntax:

run:
  experiment:
    tool: tool
    type: type
    params:
      parameter1: ${search_space_parameter}
      parameter2: fixed_value

# Or alternatively:
run:
  analyze:
    strategy: strategy
    data: data
    thresholds:
      threshold: ${search_space_parameter}

Complete Example¶

The following is a complete example of parameter optimization for a PCR reaction:

metadata:
  experiment_id: PCR_OPTIM_001
  researcher: Dr. María González
  project: pcr_optimization
  description: Optimization of PCR conditions to maximize specificity

optimize:
  # Define parameters to explore
  search_space:
    annealing_temperature: range(55, 72)      # Annealing temperature (°C)
    primer_concentration: range(0.1, 1.0)    # Primer concentration (μM)
    extension_time: choice([30, 45, 60])     # Extension time (seconds)
    mgcl2_concentration: range(1.5, 4.0)      # MgCl2 concentration (mM)

  # Optimization strategy with active learning
  strategy:
    name: ActiveLearning
    uncertainty_metric: entropy
    initial_samples: 8
    batch_size: 4

  # Objective: maximize PCR specificity
  objective:
    maximize: specificity_score

  # Budget constraints
  budget:
    max_experiments: 32
    max_time: 48h
    convergence_threshold: 0.005

  # Run PCR experiment in each iteration
  run:
    experiment:
      tool: PCR_amplifier
      type: thermal_cycling
      params:
        annealing_temp: ${annealing_temperature}
        primer_conc: ${primer_concentration}
        extension_time: ${extension_time}
        mgcl2_conc: ${mgcl2_concentration}
      output: pcr_results

  # Store optimized parameters
  output: optimized_pcr_conditions

Advanced Features¶

Multi-Fidelity Optimization¶

For complex optimization tasks with different fidelity levels:

optimize:
  search_space:
    temperature: range(25, 42)
    concentration: range(10, 100)

  strategy:
    name: BayesianOptimization
    fidelity_levels: [low, medium, high]
    fidelity_switching: adaptive

  objective:
    maximize: efficiency_score

  budget:
    max_experiments: 50
    max_time: 24h

  run:
    experiment:
      tool: CRISPR_editor
      fidelity: ${current_fidelity}
      params:
        temp: ${temperature}
        conc: ${concentration}

Multi-Objective Optimization¶

For scenarios with competing objectives:

optimize:
  search_space:
    temperature: range(25, 42)
    concentration: range(10, 100)

  strategy:
    name: MultiObjectiveBayesian
    pareto_front: true

  objective:
    maximize:
      - efficiency
      - specificity
    minimize:
      - cost
      - time

  budget:
    max_experiments: 100

  run:
    experiment:
      tool: GeneEditor
      params:
        temp: ${temperature}
        conc: ${concentration}

Integration with Other Blocks¶

The optimize block seamlessly integrates with other GFL workflow components:

design: Optimize parameters for generative design processes
analyze: Optimize analysis parameters based on results
refine: Improve candidate quality through iterative optimization
simulate: Optimize simulation parameters for accuracy

This integration enables sophisticated workflows that combine experimental optimization with generative design and analysis.

Performance Considerations¶

When using the optimize block, consider:

Computational Resources: Optimization algorithms may require significant computing power
Strategy Selection: Choose algorithms appropriate for your specific optimization task
Budget Constraints: Define realistic constraints for time and resources
Convergence Monitoring: Track optimization progress to ensure effectiveness
Result Validation: Always validate optimized parameters experimentally

Best Practices¶

Start Simple: Begin with basic optimization tasks and gradually increase complexity
Define Clear Objectives: Precise objectives lead to better optimization outcomes
Set Realistic Budgets: Include constraints that reflect experimental reality
Validate Results: Always experimentally validate optimized parameters
Iterate: Use optimization results to refine and improve subsequent iterations

The optimize block represents a powerful capability for automated experimental design, enabling researchers to harness the power of AI for scientific discovery and innovation.

Experiment Orchestration with the optimize Block¶

Introduction¶

Block Structure¶

search_space - Search Space¶

strategy - Optimization Strategy¶

objective - Optimization Objective¶

budget - Budget and Stopping Criteria¶

run - Execution Block¶

Complete Example¶

Advanced Features¶

Multi-Fidelity Optimization¶

Multi-Objective Optimization¶

Integration with Other Blocks¶

Performance Considerations¶

Best Practices¶

Experiment Orchestration with the `optimize` Block¶

`search_space` - Search Space¶

`strategy` - Optimization Strategy¶

`objective` - Optimization Objective¶

`budget` - Budget and Stopping Criteria¶

`run` - Execution Block¶