Effect weight fields should be mutually exclusive

[nebfield]

re: PGS002253 discussions

pygscatalog/pgscatalog.core/src/pgscatalog/core/lib/models.py

Lines 595 to 608 in e8abd3f

	@model_validator(mode="after")
	def check_effect_weights(self) -> Self:
	match (
	self.effect_weight,
	self.dosage_0_weight,
	self.dosage_1_weight,
	self.dosage_2_weight,
	):
	case None, None, None, None:
	raise ValueError("All effect weight fields are missing")
	case None, zero, one, two if any(x is None for x in (zero, one, two)):
	raise ValueError("Dosage missing effect weight")
	case _:
	return self

If effect_weight and dosage_0... columns are both present, a ValueError should be raised in the CatalogScoreVariant model, right @smlmbrt ?

[smlmbrt]

I think there is a subset of scoring files that don't meet this criteria @fyvon @ens-lgil?

[QuartzLibrary]

FYI: I happened to be in the process of running a scan to see if I could turn the weight-related fields into an enum for easier scoring, and these are the combinations of fields I saw (this is on the GRCh38 harmonised files):

{"dosage_0_weight", "dosage_1_weight", "dosage_2_weight", "effect_weight", "or"}: 163, 
{"dosage_0_weight", "dosage_1_weight", "dosage_2_weight"}: 4440, 
{"effect_weight", "hr"}: 91, 
{"effect_weight", "is_diplotype_false", "is_haplotype_false", "is_interaction_false", "is_recessive_false"}: 2164664, 
{"effect_weight", "is_diplotype_false", "is_haplotype_false", "or"}: 53, 
{"effect_weight", "is_diplotype_false", "is_haplotype_true", "or", "variant_type"}: 6, 
{"effect_weight", "is_diplotype_false", "or"}: 38, 
{"effect_weight", "is_diplotype_false"}: 211, 
{"effect_weight", "is_diplotype_true", "is_haplotype_false", "or", "variant_type"}: 11, 
{"effect_weight", "is_diplotype_true", "or", "variant_type"}: 15, 
{"effect_weight", "is_diplotype_true", "variant_type"}: 18, 
{"effect_weight", "is_dominant_false", "is_recessive_false"}: 5, 
{"effect_weight", "is_dominant_false", "is_recessive_true"}: 7, 
{"effect_weight", "is_dominant_true", "is_recessive_false"}: 7, 
{"effect_weight", "is_haplotype_false", "is_interaction_false", "is_recessive_false"}: 91, 
{"effect_weight", "is_haplotype_false", "is_interaction_false", "or"}: 53, 
{"effect_weight", "is_haplotype_false", "is_recessive_false"}: 2, 
{"effect_weight", "is_haplotype_false", "or", "variant_type"}: 9, 
{"effect_weight", "is_haplotype_false", "or"}: 1144040, 
{"effect_weight", "is_haplotype_false"}: 3533491, 
{"effect_weight", "is_haplotype_true", "is_interaction_false", "or", "variant_type"}: 14, 
{"effect_weight", "is_haplotype_true", "is_interaction_true", "or", "variant_type"}: 18, 
{"effect_weight", "is_haplotype_true", "is_recessive_false", "variant_type"}: 7, 
{"effect_weight", "is_haplotype_true", "is_recessive_true", "variant_type"}: 3
{"effect_weight", "is_haplotype_true", "or", "variant_type"}: 75, 
{"effect_weight", "is_haplotype_true", "variant_type"}: 44, 
{"effect_weight", "is_haplotype_true"}: 674, 
{"effect_weight", "is_interaction_false", "is_recessive_false"}: 12950927, 
{"effect_weight", "is_recessive_false"}: 8210, 
{"effect_weight", "is_recessive_true"}: 4, 
{"effect_weight", "or"}: 963251, 
{"effect_weight"}: 3156128641, 

As you can see, effect_weight is present along with the dosage-specific weights 163 times, I ran the scan again and the score ids are: 2246, 2247, and 2253.

The fields I included are: is_haplotype, is_diplotype, effect_weight, is_interaction, is_dominant, is_recessive, dosage_0_weight, dosage_1_weight, dosage_2_weight, or, hr, variant_type.

The enum

In case someone else is interested, this is what I ended up with:

pub enum Effect {
    Additive {
        effect_weight: NotNan<f64>,
        // Either OR, HR, or neither. Never both.
        or: Option<NotNan<f64>>,
        hr: Option<NotNan<f64>>,
    },

    /// Weights that are specific to different dosages of the effect_allele
    /// (e.g. {0, 1, 2} copies) can also be reported when the the contribution
    /// of the variants to the score is not encoded as additive, dominant, or
    /// recessive. In this case three columns are added corresponding to which
    /// variant weight should be applied for each dosage, where the column name
    /// is formated as dosage_#_weight where the # sign indicates the number of
    /// effect_allele copies.
    DosageSpecific {
        dosage_0_weight: NotNan<f64>,
        dosage_1_weight: NotNan<f64>,
        dosage_2_weight: NotNan<f64>,
        effect_weight: Option<NotNan<f64>>,
        or: Option<NotNan<f64>>,
    },

    /// The weight should be added to the PGS sum if there is at least 1 copy of the
    /// effect allele (e.g. it is a dominant allele).
    Dominant { effect_weight: NotNan<f64> },

    /// The weight should be added to the PGS sum only if there are 2 copies of the
    /// effect allele (e.g. it is a recessive allele).
    Recessive { effect_weight: NotNan<f64> },

    /// The weight should be multiplied with the dosage of more than one variant.
    /// Interactions are demarcated with a _x_ between entries for each of the
    /// variants present in the interaction.
    Interaction { effect_weight: NotNan<f64> },
}