When to use
Use this when the user has a set of candidate genes or proteins for a disease and needs a transparent, weighted ranking to decide which to pursue. Apply before committing screening or chemistry resources.
Not for: summarizing the evidence base for one gene (use bio-literature-review), deciding whether a whole program is worth starting (use bio-problem-selection), or designing the validation assay itself (use bio-experiment-design).
Method
- Assemble the candidate set and fix the disease context: indication, mechanism hypothesis, patient population.
- Score each target across weighted dimensions:
- Genetic/causal evidence — GWAS, Mendelian, Open Targets association, direction of effect.
- Biological rationale — pathway centrality, tissue expression (GTEx/HPA), model validation.
- Tractability — small-molecule pockets, antibody accessibility (surface/secreted), existing chemical matter, structures.
- Safety liabilities — essentiality (DepMap), broad expression, known on-target tox, knockout phenotypes.
- Novelty & competition — patent/clinical landscape, precedented vs first-in-class.
- Normalize each dimension to 0–1. Apply explicit, user-tunable weights and show the arithmetic.
- Decision: run a sensitivity check. If the ranking flips under a plausible ±20% weight change, report it as unstable rather than as a firm order.
- Decision: flag any target that ranks only via a single dimension; a lone high genetics score with zero tractability is a research target, not a drug target.
- Recommend a top tier with go/no-go rationale and the single biggest de-risking experiment per target.
Example
Five candidates for a fibrosis program. Target A scores genetics 0.9 but tractability 0.1 (intrinsically disordered, no pocket) — flagged single-dimension, demoted to tool-compound track. Target C scores 0.6/0.7/0.8/0.7/0.5; weighted total 0.66 tops the list and survives the sensitivity sweep. The de-risking experiment for C: a knockdown in primary lung fibroblasts measuring collagen deposition.
Pitfalls
- Hiding the weights so the ranking looks objective when it encodes your priors.
- Scoring novelty as "high" without actually checking the clinical/patent landscape.
- Ignoring DepMap essentiality — a pan-essential gene is a toxicity trap.
- Reporting a single ordered list with no sensitivity note, implying false precision.
Output format
# Target Prioritization — <Indication>
Weights: genetics=_, biology=_, tractability=_, safety=_, novelty=_
## Scoring table
| Target | Genetics | Biology | Tractability | Safety | Novelty | Weighted |
## Ranked shortlist
1. <Target> — <two-line rationale> | de-risking exp: <one line>
## Risk flags & missing data
- <target>: <single-dimension flag / key gap>
Sensitivity: <stable | flips under weight change X>
Cite sources; mark low-confidence scores UNVERIFIED.
Reference
Weighted scoring rubric
Score every target 0–1 on each dimension, multiply by the weight, sum to a weighted total. Default weights below reflect that human genetic causality is the strongest predictor of clinical success (Nelson 2015 / King 2019: genetically supported targets roughly double the odds of approval). Tune weights to the program and always publish them.
| Dimension |
Default weight |
0.0 anchor |
0.5 anchor |
1.0 anchor |
| Genetic / causal evidence |
0.30 |
No association |
GWAS hit, no fine-mapping/direction |
Coding Mendelian or MR-supported causal allele, direction matches therapeutic hypothesis |
| Tractability |
0.20 |
Undruggable (disordered, no pocket, intracellular for a biologic) |
Ligandable pocket predicted, no chemical matter |
Precedented modality: known pocket + tool cpds, or surface/secreted for antibody |
| Safety / tolerability |
0.20 |
Pan-essential (DepMap CERES < −0.5 broadly), severe KO phenotype |
Some broad expression, manageable |
Restricted expression, benign LoF humans (pLoF carriers healthy) |
| Biological rationale |
0.15 |
Peripheral to mechanism |
Pathway-adjacent, model-supported |
Node is causal driver, disease-tissue expressed, model-validated |
| Novelty / competition |
0.15 |
Approved drug already on target |
Multiple clinical programs |
First-in-class, clear IP space |
Worked example — fibrosis, 3 candidates (default weights):
| Target |
Gen (0.30) |
Tract (0.20) |
Safety (0.20) |
Bio (0.15) |
Novel (0.15) |
Weighted |
| A |
0.90 |
0.10 |
0.60 |
0.80 |
0.70 |
0.605 |
| B |
0.50 |
0.80 |
0.70 |
0.60 |
0.40 |
0.595 |
| C |
0.60 |
0.70 |
0.80 |
0.70 |
0.50 |
0.660 |
C leads. A is a single-dimension case (genetics 0.9, tractability 0.1) → demote to tool-compound / genetic-validation track, not a drug program. A and B are within 0.01 → report the top as a tie pending the sensitivity sweep.
Key data sources per dimension
- Genetics/causality: Open Targets Genetics (L2G score, colocalization), GWAS Catalog, ClinVar/OMIM (Mendelian), gnomAD (pLoF constraint: pLI, LOEUF — LOEUF < 0.35 = constrained/essential-like; pLoF-tolerant genes with healthy homozygous carriers are safer knockdown targets), Mendelian randomization studies.
- Tractability: Open Targets tractability buckets (SM / AB / PROTAC / other modalities, buckets 1–10), canSAR, structural coverage (PDB / AlphaFold pLDDT), DrugBank/ChEMBL for existing chemical matter, and DGIdb.
- Safety/essentiality: DepMap (Chronos/CERES gene-effect: strongly negative = fitness-essential = tox risk), GTEx/Human Protein Atlas tissue specificity (τ score — high τ = restricted = safer), mouse KO phenotypes (IMPC/MGI), known on-target adverse effects.
- Novelty/competition: patents (SureChEMBL/Espacenet), ClinicalTrials.gov, Pharmaprojects/Cortellis, company pipelines.
Sensitivity & decision rules
Run a ±20% weight perturbation and a leave-one-dimension-out check. If the top rank changes under either, report the ranking as unstable and widen the top tier. Treat a target scoring high on only one axis as a research question, not a drug target. Never present a single ordered list without the sensitivity note — it implies false precision the underlying data does not support.
1---2name: bio-target-prioritization3description: Rank candidate drug targets by genetic evidence, tractability, safety, and novelty into a transparent, weighted, sensitivity-tested shortlist.4---56## When to use78Use this when the user has a set of candidate genes or proteins for a disease and needs a transparent, weighted ranking to decide which to pursue. Apply before committing screening or chemistry resources.910**Not for:** summarizing the evidence base for one gene (use bio-literature-review), deciding whether a whole program is worth starting (use bio-problem-selection), or designing the validation assay itself (use bio-experiment-design).1112## Method13141. Assemble the candidate set and fix the disease context: indication, mechanism hypothesis, patient population.152. Score each target across weighted dimensions:16 - Genetic/causal evidence — GWAS, Mendelian, Open Targets association, direction of effect.17 - Biological rationale — pathway centrality, tissue expression (GTEx/HPA), model validation.18 - Tractability — small-molecule pockets, antibody accessibility (surface/secreted), existing chemical matter, structures.19 - Safety liabilities — essentiality (DepMap), broad expression, known on-target tox, knockout phenotypes.20 - Novelty & competition — patent/clinical landscape, precedented vs first-in-class.213. Normalize each dimension to 0–1. Apply explicit, user-tunable weights and show the arithmetic.224. **Decision:** run a sensitivity check. If the ranking flips under a plausible ±20% weight change, report it as unstable rather than as a firm order.235. **Decision:** flag any target that ranks only via a single dimension; a lone high genetics score with zero tractability is a research target, not a drug target.246. Recommend a top tier with go/no-go rationale and the single biggest de-risking experiment per target.2526## Example2728Five candidates for a fibrosis program. Target A scores genetics 0.9 but tractability 0.1 (intrinsically disordered, no pocket) — flagged single-dimension, demoted to tool-compound track. Target C scores 0.6/0.7/0.8/0.7/0.5; weighted total 0.66 tops the list and survives the sensitivity sweep. The de-risking experiment for C: a knockdown in primary lung fibroblasts measuring collagen deposition.2930## Pitfalls3132- Hiding the weights so the ranking looks objective when it encodes your priors.33- Scoring novelty as "high" without actually checking the clinical/patent landscape.34- Ignoring DepMap essentiality — a pan-essential gene is a toxicity trap.35- Reporting a single ordered list with no sensitivity note, implying false precision.3637## Output format3839```40# Target Prioritization — <Indication>41Weights: genetics=_, biology=_, tractability=_, safety=_, novelty=_4243## Scoring table44| Target | Genetics | Biology | Tractability | Safety | Novelty | Weighted |4546## Ranked shortlist471. <Target> — <two-line rationale> | de-risking exp: <one line>4849## Risk flags & missing data50- <target>: <single-dimension flag / key gap>5152Sensitivity: <stable | flips under weight change X>53```54Cite sources; mark low-confidence scores UNVERIFIED.5556## Reference5758### Weighted scoring rubric5960Score every target 0–1 on each dimension, multiply by the weight, sum to a weighted total. Default weights below reflect that human genetic causality is the strongest predictor of clinical success (Nelson 2015 / King 2019: genetically supported targets roughly double the odds of approval). Tune weights to the program and always publish them.6162| Dimension | Default weight | 0.0 anchor | 0.5 anchor | 1.0 anchor |63|-----------|:-:|-----------|-----------|-----------|64| Genetic / causal evidence | 0.30 | No association | GWAS hit, no fine-mapping/direction | Coding Mendelian or MR-supported causal allele, direction matches therapeutic hypothesis |65| Tractability | 0.20 | Undruggable (disordered, no pocket, intracellular for a biologic) | Ligandable pocket predicted, no chemical matter | Precedented modality: known pocket + tool cpds, or surface/secreted for antibody |66| Safety / tolerability | 0.20 | Pan-essential (DepMap CERES < −0.5 broadly), severe KO phenotype | Some broad expression, manageable | Restricted expression, benign LoF humans (pLoF carriers healthy) |67| Biological rationale | 0.15 | Peripheral to mechanism | Pathway-adjacent, model-supported | Node is causal driver, disease-tissue expressed, model-validated |68| Novelty / competition | 0.15 | Approved drug already on target | Multiple clinical programs | First-in-class, clear IP space |6970**Worked example — fibrosis, 3 candidates (default weights):**7172| Target | Gen (0.30) | Tract (0.20) | Safety (0.20) | Bio (0.15) | Novel (0.15) | Weighted |73|--------|:-:|:-:|:-:|:-:|:-:|:-:|74| A | 0.90 | 0.10 | 0.60 | 0.80 | 0.70 | **0.605** |75| B | 0.50 | 0.80 | 0.70 | 0.60 | 0.40 | **0.595** |76| C | 0.60 | 0.70 | 0.80 | 0.70 | 0.50 | **0.660** |7778C leads. A is a *single-dimension* case (genetics 0.9, tractability 0.1) → demote to tool-compound / genetic-validation track, not a drug program. A and B are within 0.01 → report the top as a tie pending the sensitivity sweep.7980### Key data sources per dimension8182- **Genetics/causality:** Open Targets Genetics (L2G score, colocalization), GWAS Catalog, ClinVar/OMIM (Mendelian), gnomAD (pLoF constraint: pLI, LOEUF — LOEUF < 0.35 = constrained/essential-like; pLoF-tolerant genes with healthy homozygous carriers are safer knockdown targets), Mendelian randomization studies.83- **Tractability:** Open Targets tractability buckets (SM / AB / PROTAC / other modalities, buckets 1–10), canSAR, structural coverage (PDB / AlphaFold pLDDT), DrugBank/ChEMBL for existing chemical matter, and DGIdb.84- **Safety/essentiality:** DepMap (Chronos/CERES gene-effect: strongly negative = fitness-essential = tox risk), GTEx/Human Protein Atlas tissue specificity (τ score — high τ = restricted = safer), mouse KO phenotypes (IMPC/MGI), known on-target adverse effects.85- **Novelty/competition:** patents (SureChEMBL/Espacenet), ClinicalTrials.gov, Pharmaprojects/Cortellis, company pipelines.8687### Sensitivity & decision rules8889Run a ±20% weight perturbation and a leave-one-dimension-out check. If the top rank changes under either, report the ranking as **unstable** and widen the top tier. Treat a target scoring high on only one axis as a research question, not a drug target. Never present a single ordered list without the sensitivity note — it implies false precision the underlying data does not support.
Matrixx0070