Thermodynamic Metrics Interpretation Guide

This guide helps you interpret the thermodynamic metrics in siRNAforge output files and optimize siRNA selection based on specific experimental needs.

Quick Reference Table

Metric	Optimal Range	Good Range	Poor Range	Units
GC Content	40-55%	35-60%	<35% or >65%	%
Asymmetry Score	0.7-1.0	0.6-0.8	<0.5	0-1 scale
Paired Fraction	0.5-0.7	0.4-0.8	<0.3 or >0.9	0-1 scale
MFE	-4 to -7	-2 to -8	<-10 or >0	kcal/mol
Duplex Stability ΔG	see note	see note	—	kcal/mol
Delta ΔG End (`dg_5p - dg_3p`)	+2 to +5	+1 to +6	<0	kcal/mol
Melting Temp	65-75°C	60-78°C	<50°C or >80°C	°C
Off-target Count	0-1	0-3	>5	count

Note on off_target_count: as of 0.6.0 this column counts genuine off-targets only — hits classified on-target, ortholog (same gene in another screened species) or repeat-mediated are excluded and reported in their own columns (on_target_hits, ortholog_hits, repeat_hits). A threshold carried over from an earlier version is now stricter than the author intended, because the same numeric cutoff now applies to a smaller, cleaner count.

Note on duplex_stability_dg: since 0.5.2 this is the ΔG of the full guide:passenger duplex, which scales with length — a fully paired 21mer measures -32 to -43 kcal/mol (median -39). There is no efficacy-validated window for it in siRNAforge: it is not enforced as a filter and does not feed composite_score. Compare candidates by the length-normalised duplex_stability_score within a run instead. Pre-0.5.2 values (roughly -5 to -25) came from folding the guide against itself and are not comparable.

Analyzing Your Results

Step 1: Load and Examine Data

# View top candidates sorted by composite score
head -10 your_results/sirna_design/candidates_pass.csv

# Check distribution of key metrics
csvcut -c gc_content,asymmetry_score,mfe,duplex_stability_dg,delta_dg_end,melting_temp_c \
  your_results/sirna_design/candidates_pass.csv | head -20

The awk filters below resolve columns by header name rather than by position, so they keep working when the column set changes (0.5.2 added off_target_screened).

Step 2: Identify High-Quality Candidates

Look for siRNAs that meet multiple criteria. Note: off_target_count now counts genuine off-targets only (see the note above), so <=3 here is a stricter bar than it was pre-0.6.0.

# Filter for high-quality candidates (example thresholds)
# GC 40-55%, asymmetry >=0.7, MFE -7..-4, less stable guide 5-prime end, <=3 GENUINE off-targets
awk -F',' '
NR==1 { for (i=1; i<=NF; i++) c[$i]=i; print; next }
$c["gc_content"]>=40 && $c["gc_content"]<=55 &&
$c["asymmetry_score"]>=0.7 &&
$c["mfe"]>=-7 && $c["mfe"]<=-4 &&
$c["delta_dg_end"]>=2 &&
$c["off_target_count"]<=3 { print }
' your_results/sirna_design/candidates_pass.csv

Step 3: Troubleshoot Poor Performance

If few candidates meet optimal criteria:

Low GC Content Issues

Problem: GC content consistently <35%
Solution: Consider relaxing GC minimum to 30% or target different transcript regions
Alternative: Focus on asymmetry and MFE scores instead

Poor Asymmetry Scores

Problem: Most candidates have asymmetry_score <0.6
Solution: Prioritize candidates with highest available asymmetry scores
Check: Verify end stability differences (delta_dg_end should be positive)

Overly Stable Duplexes

Problem: MFE values <-10 kcal/mol (guide secondary structure)
Solution: Consider candidates with less negative (higher) MFE values
Alternative: Test experimentally as some cell types handle stable duplexes better

Application-Specific Guidelines

For High-Efficiency Applications

When maximum knockdown is critical. off_target_count is genuine off-targets only, so <=1 is stricter than it reads pre-0.6.0 — a candidate with one ortholog hit and zero genuine off-targets now passes this filter, where it would have failed before the redefinition.

# Prioritize asymmetry and low off-targets
# High asymmetry, actually screened, very few GENUINE off-targets, good end asymmetry
awk -F',' '
NR==1 { for (i=1; i<=NF; i++) c[$i]=i; print; next }
$c["asymmetry_score"]>=0.8 &&
$c["off_target_screened"]=="True" &&
$c["off_target_count"]<=1 &&
$c["delta_dg_end"]>=2 { print }
' results.csv

For Broad Target Coverage

When targeting multiple isoforms. off_target_count excludes on-target/ortholog/repeat hits, so <=5 genuine off-targets is a stricter bar than the same number was before the redefinition.

# Balance efficiency with transcript coverage
# Moderate asymmetry, hits at least half the input transcripts, few GENUINE off-targets
awk -F',' '
NR==1 { for (i=1; i<=NF; i++) c[$i]=i; print; next }
$c["asymmetry_score"]>=0.6 &&
$c["transcript_hit_fraction"]>=0.5 &&
$c["off_target_count"]<=5 { print }
' results.csv

For Sensitive Cell Types

When working with difficult-to-transfect cells:

# Favor stability and moderate parameters
# Higher GC for stability, strong length-normalised duplex, mid-range Tm
awk -F',' '
NR==1 { for (i=1; i<=NF; i++) c[$i]=i; print; next }
$c["gc_content"]>=45 && $c["gc_content"]<=60 &&
$c["duplex_stability_score"]>=0.7 &&
$c["melting_temp_c"]>=65 && $c["melting_temp_c"]<=75 { print }
' results.csv

Experimental Validation Tips

Testing Multiple Candidates

Select 3-5 candidates with diverse metric profiles
Include positive controls with known effective siRNAs
Test dose response to optimize concentration

Metric Correlation Analysis

Monitor which metrics correlate with experimental success:

import pandas as pd
import seaborn as sns

# Load results and experimental data
results = pd.read_csv('sirna_results.csv')
experimental = pd.read_csv('knockdown_efficiency.csv')

# Merge and analyze correlations
combined = results.merge(experimental, on='id')
correlations = combined[['asymmetry_score', 'gc_content', 'mfe',
                        'delta_dg_end', 'knockdown_percent']].corr()

sns.heatmap(correlations, annot=True)

Iterative Optimization

Baseline Test: Use default siRNAforge parameters
Analyze Results: Identify which metrics correlate with success
Refine Parameters: Adjust thresholds based on your system
Validate: Test refined predictions experimentally

Common Troubleshooting

No High-Quality Candidates

Possible Causes:

Target sequence has unfavorable composition
Overly strict filtering parameters
Transcript region lacks optimal sites

Solutions:

Relax one parameter at a time (start with GC content)
Increase candidate pool size (--top-n 50)
Try different transcript isoforms
Consider alternative target regions

All Candidates Have High Off-targets

Approach:

Prioritize candidates with lowest off-target counts
Use experimental validation to test specificity
Consider tissue-specific expression of off-targets
Implement additional experimental controls

System-Specific Optimization

Different organisms/cell types may require adjusted thresholds:

Plant cells: Often tolerate higher GC content (45-65%)
Primary cells: May need more stable duplexes
Cancer cell lines: Often more permissive of various parameters
In vivo applications: Require stricter off-target criteria

Advanced Analysis

Composite Score Interpretation

The composite score integrates multiple factors. Understanding its components helps optimization:

# Estimate component contributions (example weights)
def estimate_composite_components(row):
    """Estimate how each metric contributes to composite score"""
    gc_component = score_gc_content(row['gc_content']) * 0.2
    asymmetry_component = row['asymmetry_score'] * 0.3
    structure_component = score_mfe(row['mfe']) * 0.2
    offtarget_component = score_offtargets(row['off_target_count']) * 0.3

    return {
        'gc': gc_component,
        'asymmetry': asymmetry_component,
        'structure': structure_component,
        'offtarget': offtarget_component
    }

Custom Scoring Functions

For specialized applications, implement custom scoring (see Custom Scoring Tutorial):

def custom_therapeutic_score(candidate):
    """Scoring optimized for therapeutic applications"""
    # Heavily weight safety (low off-targets)
    safety_score = 1.0 / (1.0 + candidate.off_target_count) * 0.5

    # Moderate weight on efficiency
    efficiency_score = candidate.asymmetry_score * 0.3

    # Stability for in vivo delivery
    stability_score = score_stability(candidate.gc_content, candidate.mfe) * 0.2

    return safety_score + efficiency_score + stability_score

Remember: These guidelines provide starting points. Experimental validation remains essential for confirming siRNA effectiveness in your specific system.