Drug screening and IC50/EC50 determination — from HTS hits to dose-response

High-throughput screening generates thousands of dose-response data points and thousands of opportunities to mistake noise for biology. The IC50 is the number a medicinal chemist will act on — it drives compound selection, SAR decisions, and resource allocation for the next synthesis cycle. A poorly fitted IC50 from a noisy assay wastes months. This piece covers the pipeline from plate QC to a ranked compound list you can defend in a project review.

Assay miniaturization and plate quality

HTS moves assays into 96-, 384- or 1536-well plates. Every miniaturization step amplifies variability sources that are invisible in the standard assay format: edge effects, liquid handling imprecision, reagent dispensing order, and plate-to-plate timing differences. Before screening a library, characterize the assay in the target format:

• Signal-to-background (S:B): ≥ 10 for a robust HTS assay; < 5 means you are screening in the noise.
• Signal window: (mean max − 3 × SD_max) − (mean min + 3 × SD_min). This is the pharmacologically accessible range where you can actually detect activity.
• Z-factor: the gold standard for HTS plate quality. Z' = 1 − 3(σ_p + σ_n) / |μ_p − μ_n|, where p = positive control, n = negative control. Z' ≥ 0.5 is typically required; Z' < 0.5 means the assay must be re-optimized before library screening.

Primary screen QC gates

Every plate in a primary screen needs plate-level QC before results enter the hit list:

1. Per-plate Z-factor: flag plates with Z < 0.5 for re-run; plates with Z < 0 are failed and discarded.
2. Positive control consistency: trend per-plate positive control signal; a systematic drift across a day's run indicates reagent degradation or instrument drift.
3. Edge-effect detection: compare row 1, row 16 (in 384-well), column 1 and column 24 mean to the plate interior — edge effects > 15% above background invalidate the border wells and shrink the usable plate area.
4. Dispensing error flags: wells with signal > 3 SD from the plate mean in DMSO-control columns are likely dispensing errors (bubbles, skips); mark as invalid before normalization.

Normalization and hit calling

Raw luminescence / fluorescence values are plate-dependent. Normalization to plate controls is mandatory:

• Percent inhibition = 100 × (signal_sample − median_neutral_controls) / (median_neutral_controls − median_maximum_inhibition_controls). Simple and interpretable; preferred for primary screens.
• Robust Z-score = (signal − median_plate) / (1.4826 × MAD_plate). Robust to outliers; preferred when plate controls are not fully trusted.
• B-score: row-and-column effect removal with a two-way median polish. Use for large format plates where systematic spatial gradients are visible in the heatmap.

Hit thresholds: primary hit calling at ≥ 3 × SD of neutral controls, or fixed percent-inhibition cutoff (often ≥ 50%). Apply both and report the overlap — single-criterion hit calling produces hit lists of unpredictable size.

Dose-response and IC50/EC50 fitting

The four-parameter logistic (4PL) / Hill equation is the standard fit for sigmoidal dose-response:

y = Bottom + (Top − Bottom) / (1 + (IC50/x)^HillSlope)

Practical requirements for a defensible IC50:

• Curve coverage: at least one concentration below 20% effect and at least one above 80% effect — a partial curve IC50 is an extrapolation, not a measurement. Flag compounds without a well-constrained top and bottom as "IC50 >max concentration tested" or "IC50 <min concentration tested".
• Number of concentrations: ≥ 8 concentrations in half-log steps; 10-point 3-fold dilution series is standard. Half-log (3.16-fold) steps reduce curve artifacts vs. log (10-fold) steps.
• Replicates: technical duplicate minimum for confirmation; triplicate for lead selection; n ≥ 6 for regulatory-grade IC50.
• Curve quality metrics: R², sum of squared residuals, Hill slope (compound mechanism flag if H > 2 or H < 0.5), and upper/lower asymptote constraints.
• 95% CI on IC50: report the CI, not just the point estimate. A compound with IC50 = 50 nM (CI: 40–65 nM) is meaningfully different from IC50 = 50 nM (CI: 5–500 nM).

Selectivity profiling

A compound with IC50 = 10 nM against the target of interest is useless if it has IC50 = 12 nM against an off-target kinase. Selectivity profiling requires running the dose-response pipeline against a panel of related targets and reporting a selectivity index (SI = IC50_off_target / IC50_target). The pipeline needs to handle ragged panels — not every compound reaches every target — and produce compound-level selectivity heat maps that flag promiscuous scaffolds early.

Aggregate analysis and compound ranking

Compound ranking integrates IC50, selectivity, and physicochemical properties. A pipeline that produces a table of IC50 values with no context has done the easy part. The informative output adds:

• SAR vector: fold-change in IC50 across a matched-molecular-pair series, color-mapped to the structural modification. Requires a structure-activity relationship module.
• Promiscuity alert: compounds active in ≥ 30% of targets in a broad selectivity panel are likely pan-assay interference compounds (PAINs) or colloidal aggregators.
• Curve class: classify each dose-response as complete sigmoidal (class I), partial (class II), flat active (class III), or inactive/toxic (class IV) — different classes warrant different follow-up strategies.
• Reproducibility score: IC50 coefficient of variation across independent replicates — a CV > 50% suggests the compound is a singleton hit rather than a robust lead.

Assay interference — what ruins HTS campaigns

Three interference patterns cause the majority of false positives in biochemical HTS:

1. Fluorescent compounds: excitation/emission overlap with the detection reagent produces apparent activity. Run counter-screens in a different detection modality (TR-FRET, AlphaScreen, orthogonal biochemical assay).
2. Colloidal aggregators: compounds that form colloidal particles at HTS concentrations non-specifically capture proteins. Identify with dynamic light scattering or by adding a detergent (0.01% Triton X-100) and retesting — true binders are unaffected, aggregators lose activity.
3. Reactive compounds: electrophiles that covalently modify the assay protein non-specifically. Pan-assay interference compound (PAINS) filters remove known scaffolds, but novel reactive compounds escape filters — orthogonal binding assays are the only reliable detection.

How AiLabrix fits

Drop the plate reader export (EnVision, Synergy, PHERAstar) plus the compound layout map. The pipeline computes Z-factors per plate, flags dispensing failures, normalizes, calls primary hits with Z-score and percent-inhibition criteria, fits 4PL dose-response curves for each compound, computes IC50 with 95% CI, classifies curves, generates selectivity heat maps, and ranks compounds with a configurable multi-criteria scoring function. Signed PDF with QC heatmaps, curve galleries, hit tables, and SAR trend plots — one report per campaign. [email protected].