Molecular biology: laboratory inputs, matrices, and results

Compare expression groups and inspect quality

The Expression Groups and Read QC example provides synthetic scores and a Sequence workbench. Open Paste scientific data, paste the example's MatrixToPaste output as Matrix Market, and choose Use pasted data. Select observations in rows and general matrix, preview the labels and values, then import. Choose Compare expression groups. Name the analysis and enter non-overlapping observation row numbers for each group; the interface numbers rows from 1. For the example, use treated rows 3,4 as A and control rows 1,2 as B.

Choose Welch t (unequal variances) for the explicit Student t reference, or Normal approximation (legacy method) to inspect the normal reference. Newly computed normal tails use a more accurate survival function; saved historical results keep their original values. Welch needs at least two independent observations in each group and assumes approximately normal values within each group, especially for small groups; the group variances may differ. These methods use nonnegative expression scores as stored; normalize separately if that is part of the study. Neither is a count-model RNA-seq analysis.

The first synthetic feature has Welch p about 0.071523 and 2 degrees of freedom; the old normal approximation gives about 0.000407. The difference comes from the reference distribution. The NIST two-sample t reference explains the unequal-variance method. Repeat the comparison under a different name to preserve both analyses.

Results show feature labels, group means, log2 fold change, statistic, degrees of freedom, p-value and BH adjusted p-value. The assumptions panel explains preprocessing, pseudocount and group membership. Detection filtering precedes BH correction across the full eligible feature family; the 1,000-row retained result limit and 50-row display pages do not redefine that family. A pseudocount changes fold change, not the test statistic. Old archives remain readable and explicitly identify assumptions that were not recorded.

For the read example, paste ReadsToPaste as FASTQ and import it. Choose Read quality control to see total reads/bases, Q20/Q30, mean length, GC fraction, and a cycle-quality plot with a table. The plot pages at 200 cycles and leaves gaps where quality is missing. Whole-read summaries remain constant across pages. PCA and SVD results show observation labels; only results with two calculated components receive a two-dimensional plot. The matrix's Centered PCA (two components) action is available when its shape supports two components.

Save the workbook archive to retain the native analyses. New named DE payloads bind their method and parameters into their content identity, so two analyses with equal numeric results cannot overwrite each other's assumptions. Updated DE authoring uses a versioned operation; an older runtime refuses it rather than silently choosing another method. Reviewed matrix import and explicit PCA/SVD workbench actions use the same refusal principle. Their new artifact identities retain import decisions and embedding methods even when the values coincide. Older saved artifacts remain readable without inferred provenance.

For portable explicit formula authoring, use BIO_OMICS_EMBED_V2(matrix, components, "pca") or "svd". All three arguments are required. Older runtimes do not know this function and report an error. The older BIO_OMICS_EMBED identity remains available; omitting its method keeps uncentered SVD, and older implementations can ignore its third argument.

Centered PCA now uses a bounded symmetric eigensolve when the smaller matrix axis is at most 64 and the work budget permits it. Larger problems keep the sparse solver. Either path may refuse excessive work, memory, or unresolved convergence; neither substitutes uncentered SVD.

Choose a complete research starter

Choose Life Science Analysis, then Browse examples. The preferred starters cover laboratory and bioinformatics tasks:

Starter	Baseline and scenario	What to inspect
Required Measurements — qPCR Study	Enter the four synthetic Cq measurements shown on the sheet to obtain expression 4; lowering treated-target Cq by one gives 8	Missing-measurement status, assumptions, and saved inputs
Enzyme Kinetics — Fit and Inspect	Synthetic observations fit Vmax 100 and Km 20; substrate 20 and 40 give rates 50 and about 66.6667	Observations versus fitted rates, measured-minus-fitted residuals, and convergence errors
Sequence Study — Translate and Inspect	The toy sequence translates to MAEF; an exact alignment target has eight matches and no mismatches	Peptide letters, ORF coordinates, the alignment summary, and the Sequence workspace
HBB Reference — Check the Coding Frame	NM_000518.5 CDS translates to the annotated 147-residue NP_000509.1 peptide; frame 2 stops after 3 residues	Versioned public data, CDS coordinates, peptide letters and a saved reading-frame scenario
Expression Groups and Read QC	Compare the two small-sample reference distributions; change formula components from two to one	Named native analyses, explicit method assumptions, and a read fixture with Q20/Q30 = 75%
Matrix Axes and Empty Observations	Import observations in columns; totals 10, 20, 0 become 10,000, 10,000, 0	Labels, the retained empty observation, and a formula target-total scenario of 100
Variant Review — Passed and Unfiltered	Baseline selects one passed call at ALT frequency 0.75; a quality threshold of 70 selects none	Passed, failed and unfiltered states, a Variant workspace, and input persistence

The qPCR starter deliberately begins with missing measurements. Its worksheet contains the synthetic values to enter in I/O; opening a demonstration is not equivalent to supplying measurements. The kinetics fit uses positive log parameters and unweighted least squares. Its residuals describe agreement with five synthetic observations; they are not parameter uncertainty or assay validation.

The matrix starter also teaches table shape: BYROW(ObservationTable, (row) => SUM(row)) returns three row totals (10, 20, 0), while MAP(ObservationTable, (count) => count * 2) transforms six entries and keeps the 3×2 shape. A MAP lambda receives an entry; use BYROW when the calculation needs a whole row.

Each starter contains editable inputs, declared results, assumptions, and a scenario guide. Save the baseline before changing a scenario. Reopening that archive restores its saved inputs and outputs; resetting inputs restores the authored defaults, including blank Cq measurements. In the kinetics starter, an iteration budget of one deliberately fails: inspect LogParameters for the failure class and recovery advice, then restore 200. A failed fit does not show a numerical prediction.

In the sequence starter, select Protein, OpenReadingFrames, or Alignment in I/O to open the inspector. Entering invalid DNA, such as ATGZ, retains the alphabet error through translation and ORF inspection. The Sequence_1 tab offers the existing file-import/artifact workflow. Import a FASTA file there before running artifact operations; the inline toy formula results are not automatically imported into that workspace.

Life-science setup now asks for a biology model and editable study inputs. Its model step requires data and the authored biology tag; generic welcome arithmetic does not complete it. Setup completion means those two conditions exist, not that someone reviewed the experiment. Creating or promoting an App is optional. Stored explicit view-mode choices are preserved.

Review variant filter status

The variant review example separates passed, failed, and unfiltered synthetic calls. At minimum quality 30, only passed is selected and its alternate-allele frequency is 0.75. Raising the threshold to 70 selects none, including the unfiltered record with quality 99.

Use variant.filter_status = "passed" for a PASS-only rule. An empty filters list alone cannot establish that filtering was applied. The Variant workspace shows filter status and offers Passed only for its current bounded window. Older records without retained status display Unknown; recover their status from original VCF data rather than guessing PASS.

Variant and alternate indexes in BIO_VARIANT_AF are one-based. Invalid ALT queries and genotype allele indexes produce errors. Missing called alleles do not enter the denominator; a valid allele with no calls returns blank, while a valid allele with called references but no alternates returns zero.

Scientific .gridoc imports verify analysis Frames against their artifact and payload identities. This supports restoring earlier content-addressed analysis Frames without changing their scientific data or bypassing integrity checks.

Use unambiguous scientific names

Use Contig_N50 for a descriptive name. ContigN50 matches Grid's cell-address syntax and exceeds the range whose identity Grid can preserve. New source admission rejects such addresses with GRID_ADDRESS_UNREPRESENTABLE and a source location, rather than mapping them to another cell.

Explicit cell addresses must be within columns A–XFC (1–16,383) and rows 1–67,108,863. Some views and imports have lower limits. XFD is not currently representable by Grid's address format, even though it is a familiar Excel column. This check does not rewrite an old archive or recover values already overwritten by an address collision. Review and rename ambiguous bindings in the retained source before using the corrected model.

Start with required laboratory inputs

The guarded laboratory example combines dilution, relative qPCR, and Michaelis–Menten initial-rate calculations. Concentrations are in micromolar and volumes in microliters. Its bounds are declared teaching windows, not universal assay-quality criteria.

Cq inputs deliberately start blank. Enter treated target 23, treated reference 20, control target 25, and control reference 20 to reproduce the synthetic relative expression of 4. Changing treated target to 22 produces 8. This comparison assumes equal target/reference efficiencies near doubling; technical replicates do not establish biological replication or assay suitability.

Clearing a Grid input restores its authored default. Because these Cq defaults are blank, clearing any measurement displays Missing Cq measurement and removes the numeric expression result. Resetting the example also restores this missing state. A stored override equal to a source default remains an override until reset.

With stock 100, target 10, and final volume 200, the dilution requires 20 microliters of stock and 180 microliters of diluent. Invalid input writes are rejected with context. Reducing stock to 5 after setting target to 10 makes the existing scenario infeasible; the example hides its volume outputs and says what needs correction. Negative substrate and unsuitable slope inputs are rejected. Zero substrate is valid and gives zero initial rate. The same source contracts apply through the interface and model API.

Review a Matrix Market file before import

In the Sequence workspace, choose a .mtx file or paste Matrix Market text, then select whether its observations (cells or samples) occupy rows or columns. Choose Raw counts for nonnegative whole numbers or General matrix for signed/transformed values. Optional observation and feature labels contain one identifier per line, in the chosen axis order. Label counts must match the dimensions and identifiers must be unique within each axis.

Preview matrix displays the interpreted dimensions, labels, values, and observation totals before Import reviewed matrix becomes available. Changing the file, axes, values contract, or labels requires another preview. No normalization or filtering occurs during import. Grid retains the chosen orientation and transpose in the imported artifact alongside its labels. New imports include those decisions in the content identity, so an equivalent matrix imported under another declared value type cannot overwrite its history.

Pasted data uses the same review and native import as file data. It is not submitted when Use pasted data is clicked: matrix data still needs preview and import; FASTA, FASTQ and VCF use the normal Import button. Copy the actual output text, including newlines and tabs, rather than the quoted formula source.

The reviewed import accepts UTF-8 Matrix Market coordinate real/integer general files up to 60 MiB. A label is at most 1,024 bytes, with at most 16 MiB of labels per axis. Existing sparse matrix dimension and structural limits also apply. Unknown orientation/value choices, mismatched labels, unsupported encodings, and invalid raw counts produce an error.

The matrix view shows at most 50 observations and eight features, with totals over all features. Import review shows five observations. For the retained three-observation example, totals 10, 20, and 0 become 10,000, 10,000, and 0 after Normalize totals to 10,000 (no log). Zero-total observations remain visible. The operation creates another artifact so the imported data stays available. Use the centered PCA or uncentered SVD choice explicitly for the next analysis.

Linked Matrix Market assets retain the older observations-in-rows convention; use the file import when you need axis review or transposition. Earlier imports without explicit options keep their previous parsing behavior. Their original axis choice cannot be reconstructed from shape alone.

Read the result without decoding its storage

The cell inspector and expanded cell show protein letters for tagged translations, an ORF table, and a common alignment summary for the existing alignment field conventions. ORF coordinates use zero-based, end-exclusive positions on the original forward strand, including negative reading frames. Alignment query and target intervals use the same interval convention.

Previews are deliberately bounded: up to 2,000 translation letters, 50 ORFs with 120 letters each, and 100 structured CIGAR operations. Truncation is labeled. Expanded cells retain access to the original fields. The Sequence workspace uses bounded artifact projections for matrices, ORFs, and alignments.

I/O keeps computation status separate from worksheet placement. A computed array can be ready while a Placement blocked notice identifies an obstruction and offers a jump to it. Removing the obstruction clears the notice when the updated placement arrives. Cell-address outputs use natural column/row order (A1, A2, A10); named output slots and explicit saved ordering are preserved.

Centered PCA or uncentered SVD

Choose the mathematical method explicitly when comparing observation scores. Grid stores sparse count matrices with observations in rows and features in columns. Verify those axes before analysis; a transposed matrix describes a different problem.

Centered = BIO_OMICS_EMBED(Counts, 1, "pca")
Uncentered = BIO_OMICS_EMBED(Counts, 1, "svd")

PCA subtracts each feature's mean across observations. It does not scale features to unit variance. SVD uses the uncentered values. The selected method is part of the model source and survives saving. An omitted method preserves Grid's historical uncentered SVD result; existing models and artifacts are not automatically migrated.

For observation rows [80, 20], [20, 80], and [0, 0], the first scores are:

Observation	Centered PCA	Uncentered SVD
First	-42.426407	70.710678
Second	42.426407	70.710678
Empty	0	0

Each column may have its sign reversed without changing its meaning. Adding the same offset to every observation of a feature preserves centered PCA scores. Tied component variances can produce different valid bases, so compare the represented subspace or pairwise geometry in that case.

Centered PCA needs at least two observations and between one and min(features, observations - 1) components. Constant data and numerically null remaining directions have zero scores. Sparse power iteration is bounded by 512 iterations per component and 128 MiB of auxiliary memory. The operation reports invalid inputs, excessive work, numerical failure, or nonconvergence; it never switches to uncentered SVD to produce an answer.

Use a runtime containing the explicit method option. Older runtimes did not interpret the third argument. Earlier scientific artifacts labeled sparse_pca used the uncentered operation. New native embedding artifacts record the method, centering, scaling, and algorithm version. Existing omics_cluster analysis retains its uncentered embedding.

The complete embedding example includes the matrix, translated observations, and expected values. It is a method demonstration, not a complete single-cell analysis protocol.

Count and locate DNA motifs

Matches = DNA_MOTIF_COUNT("AAAAA", "AAA")
Position = DNA_FIND_MOTIF("CCATG", "ATG")

Both results are 3. The count includes overlapping matches starting at positions 1, 2, and 3. The position is 1-based; an absent motif returns 0. Matching ignores ASCII case and compares the literal sequence, without expanding ambiguity codes into wildcard matches. Pass sequence and motif strings. A numeric motif produces an input error with a valid example.

These helpers do not validate a complete biological sequence. Use BIO_SEQUENCE("ATGZ", "dna") to see the packed-sequence alphabet diagnostic for an invalid symbol. Existing empty-string behavior is retained: DNA_MOTIF_COUNT(sequence, "") returns 0 and DNA_FIND_MOTIF(sequence, "") returns 1.