aug26-update

DOCS.md

@@ -1,68 +1,205 @@
-# Demo Instance Lab Dataset
+---
+title: "Binding Scores"
+description: "Schema reference for computed binding score tables"
+---
+
+These are the computed outputs of the data labeling pipeline.
+
+---
+
+## Binding Scores
 
-This repo contains three example datasets that would be available to the user for downstream model training. The derived data is calculated from log probabilities of each library candidate in an assay sample. The different data modalities are:
+**File:** `binding-scores-YYYYMMDD.parquet`
+
+Aggregated binding scores per candidate and antigen pair. Replicate binding scores are averaged across replicates and pivoted so that each antigen concentration becomes its own column.
 
-- Expression
-- Binding
-- Specificity
+**Grain:** one row per unique `(candidate_id, sino_catalog_id)`.
 
-Expression scores are log enrichments of the candidate translated protein frequency relative to the initial DNA pool candidate frequency. The binding score is defined as the ratio of the relative enrichment of a candidate over the enrichment of the weakest candidate in the dataset represented on a log scale. The specificity score is defined in terms of binding scores of off-target and on-target antigens, and is the ratio of the targeting binding score divided by the average to the binding scores of all off-target interactions.
+### Fixed Columns
 
-## Common Metadata
+| Column | Type | Description |
+| :--- | :--- | :--- |
+| `candidate_library_id` | `string` | Identifier of the candidate library consumed by this sample. |
+| `candidate_name` | `string` | Customer-provided sequence name. |
+| `candidate_id` | `string` | Blake-3 content digest of the reference sequence. |
+| `candidate_sequence` | `string` | The candidate amino acid sequence. |
+| `variant_id` | `string` | Blake-3 content digest of the variant sequence. |
+| `variant_sequence` | `string` | The variant sequence. |
+| `sino_catalog_id` | `string` | Sino catalog identifier for the antigen used in the binding assay. |
+| `antigen_gene_name` | `string` | Gene name of the antigen. |
+| `antigen_sequence` | `string` | Amino acid sequence of the antigen. |
+| `targeting` | `boolean` | Whether the candidate was designed to target this antigen. |
+| `is_expressed` | `boolean` | Whether the candidate is expressed in at least one sample. |
 
-There are columns that are common to each of the three datasets, which facilitate simple associations if one wants to compute features over multiple different scores. The columns are:
+### Dynamic Concentration Columns
 
-- `candidate_library_id`: This ID refers to the specific sample that contains all the translated proteins expressed together. In binding and specificity contexts, this ID represents pool of binder proteins that is mixed with different antigens to generate binding and specificity scores.
-- `candidate_name`: The user-provided name of the candidate sequence.
-- `candidate_sequence`: The amino acid sequence of the candidate binder.
-- `candidate_id`: A cryptographic hash of the candidate sequence.
-- `candidate_type`: An enum field representing the type of candidate protein. In this demo, it's either `miniprotein` or `vhh`.
-- `target_pdb_id`: The PDB ID of the target protein that conditioned the model to generate the candidate sequence.
+<Note>
+In addition to the fixed columns above, this table contains **dynamic columns** for each antigen concentration used in the experiment. The column names follow the pattern `binding_score_{conc}nM` and `binding_score_{conc}nM_std`.
+</Note>
 
-## Expression Dataset
+For our standard workflow, you will see these additional columns:
 
-The unique identifiers for each row in this dataset is `["candidate_library_id", "candidate_id"]`. So, the number of rows are equal to the number of candidate proteins in each distinct pool. The scoring columns are:
+| Column | Type | Description |
+| :--- | :--- | :--- |
+| `binding_score_5nM` | `float` | Mean log binding score at 5 nM. |
+| `binding_score_5nM_std` | `float` | Standard error of the log binding score at 5 nM. |
+| `binding_score_50nM` | `float` | Mean log binding score at 50 nM. |
+| `binding_score_50nM_std` | `float` | Standard error of the log binding score at 50 nM. |
+| `binding_score_500nM` | `float` | Mean log binding score at 500 nM. |
+| `binding_score_500nM_std` | `float` | Standard error of the log binding score at 500 nM. |
+
+---
+
+## Replicate Binding Scores
+
+**File:** `replicate-binding-scores-YYYYMMDD.parquet`
+
+Per-replicate binding scores unrolled across antigen concentrations. Each row represents a single candidate measured in a specific expression/binding sample pair at a specific antigen concentration. Use this table to inspect replicate-level variation and per-sample UMI counts.
+
+**Grain:** one row per `(candidate_id, binding_sample_id, antigen_concentration_nM)`.
+
+| Column | Type | Nullable | Description |
+| :--- | :--- | :--- | :--- |
+| `candidate_library_id` | `string` | no | Identifier of the candidate library consumed by this sample. |
+| `expression_sample_id` | `string` | no | Sample identifier for the expression (base) sample. |
+| `binding_sample_id` | `string` | no | Sample identifier for the binding (dose) sample. |
+| `candidate_name` | `string` | no | Customer-provided sequence name. |
+| `candidate_id` | `string` | no | Blake-3 content digest of the reference sequence. |
+| `candidate_sequence` | `string` | no | The candidate amino acid sequence. |
+| `variant_id` | `string` | no | Blake-3 content digest of the variant sequence. |
+| `variant_sequence` | `string` | no | The variant sequence. |
+| `sino_catalog_id` | `string` | no | Sino catalog identifier for the antigen. |
+| `antigen_gene_name` | `string` | no | Gene name of the antigen. |
+| `antigen_sequence` | `string` | no | Amino acid sequence of the antigen. |
+| `antigen_concentration_nM` | `float` | no | Antigen concentration in nanomolar. |
+| `targeting` | `boolean` | no | Whether the candidate was designed to target this antigen. |
+| `is_expressed` | `boolean` | no | Whether the candidate is expressed. |
+| `binding_score` | `float` | yes | Log binding score. Null when UMI count thresholds are not met. |
+| `expression_umi_count` | `integer` | yes | UMI count from the expression (base) sample. |
+| `binding_umi_count` | `integer` | yes | UMI count from the binding (dose) sample. |
+
+**Key relationships:**
+- `expression_sample_id` and `binding_sample_id` both correspond to `dim_samples.sample_id`.
+- `candidate_id` is the shared candidate key across all tables.
+- `variant_id` is the shared variant key across all tables.
 
-- `expression_score`: The log enrichment of the expression of the translated candidate relative to the double stranded pool.
-- `expression_score_std`: The standard error of the expression score estimate calculated from each of the replicates measured.
+## Specificity Scores
 
-## Binding Dataset
+**File:** `specificity-scores-YYYYMMDD.parquet`
 
-The unique identifiers for each row in this dataset is `["candidate_library_id", "candidate_id", "antigen_pdb_id"]`. So, the number of rows are equal to the number of candidate protein + antigen combinations in each distinct library. The columns specific to the binding dataset are:
+Aggregated specificity scores per candidate and antigen pair. Replicate specificity scores are averaged across replicates and pivoted so that each antigen concentration becomes its own column.
 
-- `antigen_pdb_id`: The PDB ID of the antigen protein used in the binding assay.
-- `antigen_gene_name`: The gene name associated with the PDB ID of the antigen.
-- `sino_catalog_id`: The ID of the antigen ordered from Sino Biological.
-- `antigen_sequence`: The expressed sequence of the antigen used in the assay.
-- `targeting`: Boolean value that specifies whether the antigen used in the binding assay matches the target for the candidate protein.
-- `binding_score_50nM`: The mean binding score from the assay where the candidate library was tested with 50 nM of the antigen protein.
-- `binding_score_500nM`: The mean binding score from the assay where the candidate library was tested with 500 nM of the antigen protein.
-- `binding_score_50nM_std`: The standard error of the mean 50 nM binding score estimate calculated from each of the replicates measured.
-- `binding_score_500nM_std`: The standard error of the mean 500 nM binding score estimate calculated from each of the replicates measured.
+**Grain:** one row per unique `(candidate_id, sino_catalog_id)`.
 
-## Specificity Dataset
+### Fixed Columns
 
-The unique identifiers for each row in this dataset is `["candidate_library_id", "candidate_id"]`. So, the number of rows are equal to the number of candidate proteins in each distinct pool. This columns specific to the specificity dataset are:
+| Column | Type | Description |
+| :--- | :--- | :--- |
+| `candidate_library_id` | `string` | Identifier of the candidate library consumed by this sample. |
+| `candidate_name` | `string` | Customer-provided sequence name. |
+| `candidate_id` | `string` | Blake-3 content digest of the reference sequence. |
+| `candidate_sequence` | `string` | The candidate amino acid sequence. |
+| `variant_id` | `string` | Blake-3 content digest of the variant sequence. |
+| `variant_sequence` | `string` | The variant sequence. |
+| `sino_catalog_id` | `string` | Sino catalog identifier for the antigen used in the binding assay. |
+| `antigen_gene_name` | `string` | Gene name of the antigen. |
+| `antigen_sequence` | `string` | Amino acid sequence of the antigen. |
+| `is_expressed` | `boolean` | Whether the candidate is expressed in at least one sample. |
+
+### Dynamic Concentration Columns
+
+<Note>
+In addition to the fixed columns above, this table contains **dynamic columns** for each antigen concentration used in the experiment. The column names follow the pattern `log_specificity_score_{conc}nM` and `log_specificity_score_{conc}nM_std`.
+</Note>
+
+For our standard workflow, you will see these additional columns:
+
+| Column | Type | Description |
+| :--- | :--- | :--- |
+| `log_specificity_score_5nM` | `float` | Mean log specificity score at 5 nM. |
+| `log_specificity_score_5nM_std` | `float` | Standard error of the log specificity score at 5 nM. |
+| `log_specificity_score_50nM` | `float` | Mean log specificity score at 50 nM. |
+| `log_specificity_score_50nM_std` | `float` | Standard error of the log specificity score at 50 nM. |
+| `log_specificity_score_500nM` | `float` | Mean log specificity score at 500 nM. |
+| `log_specificity_score_500nM_std` | `float` | Standard error of the log specificity score at 500 nM. |
+
+---
+
+## Replicate Specificity Scores
+
+**File:** `replicate-specificity-scores-YYYYMMDD.parquet`
+
+Per-replicate specificity scores unrolled across antigen concentrations. Each row represents a single candidate measured in a specific expression/binding sample pair at a specific antigen concentration. Use this table to inspect replicate-level variation.
+
+**Grain:** one row per `(candidate_id, binding_sample_id, antigen_concentration_nM)`.
+
+| Column | Type | Nullable | Description |
+| :--- | :--- | :--- | :--- |
+| `candidate_library_id` | `string` | no | Identifier of the candidate library consumed by this sample. |
+| `expression_sample_id` | `string` | no | Sample identifier for the expression (base) sample. |
+| `binding_sample_id` | `string` | no | Sample identifier for the binding (dose) sample. |
+| `candidate_name` | `string` | no | Customer-provided sequence name. |
+| `candidate_id` | `string` | no | Blake-3 content digest of the reference sequence. |
+| `candidate_sequence` | `string` | no | The candidate amino acid sequence. |
+| `variant_id` | `string` | no | Blake-3 content digest of the variant sequence. |
+| `variant_sequence` | `string` | no | The variant sequence. |
+| `sino_catalog_id` | `string` | no | Sino catalog identifier for the antigen. |
+| `antigen_gene_name` | `string` | no | Gene name of the antigen. |
+| `antigen_sequence` | `string` | no | Amino acid sequence of the antigen. |
+| `antigen_concentration_nM` | `float` | no | Antigen concentration in nanomolar. |
+| `is_expressed` | `boolean` | no | Whether the candidate is expressed. |
+| `log_specificity_score` | `float` | yes | Log specificity score. Null when UMI count thresholds are not met. |
+
+**Key relationships:**
+- `expression_sample_id` and `binding_sample_id` both correspond to `dim_samples.sample_id`.
+- `candidate_id` is the shared candidate key across all tables.
+- `variant_id` is the shared variant key across all tables.
+
+## Expression Scores
+
+**File:** `expression-scores-YYYYMMDD.parquet`
+
+Aggregated expression scores per candidate. Expression scores are averaged across replicates.
+
+**Grain:** one row per unique `(candidate_id, variant_id)`.
+
+### Fixed Columns
+
+| Column | Type | Description |
+| :--- | :--- | :--- |
+| `candidate_library_id` | `string` | Identifier of the candidate library consumed by this sample. |
+| `candidate_name` | `string` | Customer-provided sequence name. |
+| `candidate_id` | `string` | Blake-3 content digest of the reference sequence. |
+| `candidate_sequence` | `string` | The candidate amino acid sequence. |
+| `variant_id` | `string` | Blake-3 content digest of the variant sequence. |
+| `variant_sequence` | `string` | The variant sequence. |
+| `is_expressed` | `boolean` | Whether the candidate is expressed in at least one sample. |
+| `expression_score` | `float` | Mean expression score across all samples. |
+| `expression_score_std` | `float` | Standard error of the expression score across all samples. |
 
-- `target_gene_name`: The gene name associated with the PDB ID of the target antigen.
-- `sino_catalog_id`: The ID of the target antigen ordered from Sino Biological.
-- `target_sequence`: The expressed sequence of the target antigen used in the assay.
-- `log_specificity_score_50nM`: The mean specificity score from the assay where the candidate library was tested with 50 nM of the antigen protein.
-- `log_specificity_score_500nM`: The mean specificity score from the assay where the candidate library was tested with 500 nM of the antigen protein.
-- `log_specificity_score_50nM_std`: The standard error of the mean 50 nM specificity score estimate calculated from each of the replicates measured.
-- `log_specificity_score_500nM_std`: The standard error of the mean 500 nM specificity score estimate calculated from each of the replicates measured.
+---
+
+## Replicate Expression Scores
 
-## Replicate Datasets
+**File:** `replicate-expression-scores-YYYYMMDD.parquet`
 
-We include datasets that contain all of the replicates that were used to generate the binding and specificity mean scores/standard errors. As a result, the standard error columns are removed, and, for simplicity, we provide a row for each dose (column called `antigen_concentration_nM`). To further improve our data provenance, we include unique molecular identifier (UMI) counts for the expression and binding samples, along with sample ids for the expression and binding samples. The added columns are called:
+Per-replicate expression scores unrolled across different biological replicates. Use this table to inspect replicate-level variation.
 
-- `expression_sample_id`: The sample id of the expression sample. Different ID's that map to the same `candidate_library_id` are repeated measurements of the same expressed pool (which is the candidate_library_id). This means any variation observed here is likely measurement variation and not due to biological variation.
-- `binding_sample_id`: The sample id of the binding sample. Different ID's of this that map to the same `candidate_library_id` are separate biological replicates of the same pool as they represent distinct binding attempts. This means that any variation observed along this axis is due to a combination of measurement and biological variation.
-- `expression_umi_count`: The number of unique molecular identifiers (UMIs) observed for the binder in the expression sample.
-- `binding_umi_count`: The number of unique molecular identifiers (UMIs) observed for the binder in the binding sample.
-
-### Notes
-
-When estimating our binding and specificity scores, we use a weak Bayesian prior and the UMI counts update the posterior. This means that zero counts are not truly zero, just that the probability of seeing that candidate is likely lower than one over the number of observations. 
-
-These extra columns are included in the binding replicates dataset, while the other two (expression-replicates and specificity-replicates) do not, for now. They can be made available upon request!
+**Grain:** one row per `(expression_sample_id, candidate_id, variant_id)`.
+
+| Column | Type | Nullable | Description |
+| :--- | :--- | :--- | :--- |
+| `candidate_library_id` | `string` | no | Identifier of the candidate library consumed by this sample. |
+| `expression_sample_id` | `string` | no | Sample identifier for the expression (base) sample. |
+| `candidate_name` | `string` | no | Customer-provided sequence name. |
+| `candidate_id` | `string` | no | Blake-3 content digest of the reference sequence. |
+| `candidate_sequence` | `string` | no | The candidate amino acid sequence. |
+| `variant_id` | `string` | no | Blake-3 content digest of the variant sequence. |
+| `variant_sequence` | `string` | no | The variant sequence. |
+| `is_expressed` | `boolean` | no | Whether the candidate is expressed. |
+| `expression_score` | `float` | yes | Expression score derived from the log enrichment from synthesis to expression. Null when UMI count thresholds are not met. |
+
+**Key relationships:**
+- `expression_sample_id` corresponds to `dim_samples.sample_id`.
+- `candidate_id` is the shared candidate key across all tables.
+- `variant_id` is the shared variant key across all tables.

README.md

@@ -1,28 +1,28 @@
 ---
 configs:
-- config_name: expression
+- config_name: expression_scores
   data_files:
   - split: train
-    path: "expression/expression-scores.parquet"
-- config_name: expression-replicates
+    path: "expression_scores/**/*.parquet"
+- config_name: replicate_expression_scores
   data_files:
   - split: train
-    path: "expression-replicates/expression-scores-with-replicates.parquet"
-- config_name: binding
+    path: "replicate_expression_scores/**/*.parquet"
+- config_name: binding_scores
   data_files:
   - split: train
-    path: "binding/binding-scores.parquet"
-- config_name: binding-replicates
+    path: "binding_scores/**/*.parquet"
+- config_name: replicate_binding_scores
   data_files:
   - split: train
-    path: "binding-replicates/binding-scores-with-replicates.parquet"
-- config_name: specificity
+    path: "replicate_binding_scores/**/*.parquet"
+- config_name: specificity_scores
   data_files:
   - split: train
-    path: "specificity/specificity-scores.parquet"
-- config_name: specificity-replicates
+    path: "specificity_scores/**/*.parquet"
+- config_name: replicate_specificity_scores
   data_files:
   - split: train
-    path: "specificity-replicates/specificity-scores-with-replicates.parquet"
+    path: "replicate_specificity_scores/**/*.parquet"
 license: mit
 ---

binding/binding-scores.parquet → binding_scores/binding-scores-260812.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93d297948aaf13e5ea2669eafd8b455a23765ef5fc80dd52392178b8881460dd
-size 4601677
+oid sha256:bf4604d2b74d4407d314085eef47453507c0dd41b4f8f80071e0531aae71957c
+size 3893628

specificity-replicates/specificity-scores-with-replicates.parquet → replicate_binding_scores/replicate-binding-scores-260812.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:26e81858c6fa80d5309847710c0ce329aebf145be127a00fb69b470afe7b3235
-size 3819349
+oid sha256:9a8cac744df7b5474b67a2b241f2d7b851fda979091f67a6e15085283ea179ea
+size 5706560

specificity/specificity-scores.parquet → replicate_specificity_scores/replicate-specificity-scores-260812.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a811688af509313c0464845f4e1dfa662375cf149a2c05a1d288718862607431
-size 1819210
+oid sha256:eb74ba9a661df8dfcca20897bd7bfd83c87689eaeb2bcf566791498c69792609
+size 1249166

binding-replicates/binding-scores-with-replicates.parquet → specificity_scores/specificity-scores-260812.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:eb28d3361aecdec4371f11b0349777136c5ecdf4b68531b98efc0fc70f83aed1
-size 12354226
+oid sha256:4cc30e988c04517c3c093498dbf1360f3f9032965974f3aaac6dfce2e66c83bc
+size 938652