Platform · Educational guide

Data analysis foundations

How to read sample tables, recognize biological-data formats, and interpret recorded values before choosing an analysis method.

This guide supports the BioDev and machine-learning learning pathways and prepares visitors to study methods proposed for the Virtual Lab. It does not execute analyses.

Português (Brasil)

Analysis foundations

Data for analysis: types, formats, and examples

A single study can combine sample tables, DNA sequences, and leaf photographs. Record organization determines how to read these files; each variable’s measurement scale determines how its values can be compared.

Data, information, evidence, and knowledge

Data are records of characteristics of events, objects, or phenomena: measurements, counts, observations, text, and images. They are basic inputs to an information system, which brings together people, processes, and resources to record, organize, and interpret them. A dataset is a collection of such records with a defined purpose and origin.

  1. Data: the fictional record “A01; dry mass; 2 g” describes a sample and a measurement.
  2. Information: comparing samples measured using the same procedure shows how their masses differ. A number without a unit and method conveys little.
  3. Knowledge: in science, explanations supported by evidence and critical evaluation help us understand phenomena and guide further investigation, while remaining open to revision. Processing data does not, by itself, guarantee a valid conclusion.

A hypothesis is a proposed explanation to be evaluated. Evidence is the role that data and results play in supporting or challenging a claim, considering methods, context, and limitations. In the fictional example, mass differences describe the samples; they do not demonstrate that temperature caused those differences. Moving from records to explanations requires justification, not just processing. [20]

Data can be strategic assets: access, control, and analytical capacity influence research, economies, and geopolitical relations. Their value depends on the question, quality, and conditions of use; accumulating records does not replace governance. At Kannabium, retaining sources, versions, and context makes interpretations reviewable without losing the original record.

For further study of data analysis and machine learning, see the works by Faceli and colleagues and by Fávero and Belfiore, listed as further reading. [14] [15]

Characterize before analyzing

The same dataset can be described using several criteria. They complement one another rather than forming competing categories.

Structure
How records are organized: a table with fixed fields, a document with flexible fields, or content such as images and free text.
Time and frequency
A snapshot, repeated measurements, or a continuous stream. Record dates, intervals, and gaps: not every series has equally spaced observations.
Dependence
Repeated measurements of one plant, related individuals, and nearby locations can be associated. Treating them as independent observations can distort an analysis.
Volume
Consider record and attribute counts, file sizes, and available resources. A small table and a genome collection require different storage and processing strategies.

Big Data: digital transformation, architecture, and analysis

Big Data is a phenomenon of digital transformation marked by the massive production and circulation of data across networks, services, and connected systems. In technical terms, it describes datasets whose characteristics — volume, variety, velocity, and/or variability — require a scalable architecture and suitable technologies for efficient storage, processing, and analysis. The concept therefore connects data challenges with the technological capacity needed to address them. [16]

A scalable architecture can increase its capacity as demand grows. This may mean adding resources or distributing storage and processing across several machines. This approach combines technologies for collection, integration, storage, and analysis, selected according to the operations, response time, and available resources; there is no universal cutoff in rows or gigabytes. More data do not guarantee useful information either: quality, context, and methods remain essential.

Classification example: integrating genomic sequences, images, and sensor series from many samples may require this approach. A file in FASTA (a text format for sequences) or several different formats, on their own, do not make a dataset Big Data. The workload and the capacity to process it must be assessed.

These data can come from Web access logs, e-commerce, social media, documents, sensors, radio-frequency identification (RFID), and satellite positioning (GPS). They are also produced in astronomy, atmospheric science, genomics, biochemical experiments, and clinical, telephone, image, and video records. Each source brings its own requirements for integration, processing, and interpretation.

  • Volume: the amount of data generated, collected, and retained. At massive scale, storing and analyzing these records may require increased capacity, distributed processing, and specialized tools, depending on the workload. Planning considers records, attributes, file sizes, and expected growth.
  • Velocity: the pace of data generation, collection, and processing relative to the time available to use them. Social media and monitoring systems can produce continuous streams; infrastructure must keep up with incoming data and deliver results within the time required by the application.
  • Variety: the diversity of sources, structures, and formats, such as spreadsheets, text, images, audio, video, and biological sequences. Integration requires understanding each record and aligning identifiers, units, and metadata while preserving differences relevant to the analysis.
  • Value: the usefulness of data for a question, an investigation, or a decision. Records do not all contribute equally. Analysis seeks to select and combine what is relevant to produce useful information; accumulated quantity alone does not represent a benefit.
  • Veracity: concerns the reliability and authenticity of records and their origins. It requires assessing errors, gaps, bias, and possible manipulation, and documenting methods and transformations. Traceability helps verify interpretations and communicate uncertainty, reducing the risk of misleading conclusions.
  • Variability: changes in the rate, distribution, or meaning of records over time.
  • Complexity: the difficulty of integrating different sources, relationships, units, and contexts.

The dimensions complement one another: volume, velocity, and variety guide technological planning; value and veracity help assess usefulness and quality. Variability and complexity add the challenges of change and integration. Lists of “Vs” vary between authors and are not an automatic classification test.

Record organization

Structured

Follow a defined model with standardized fields and value types. [1]

Example: a table in which each row represents a sample and the columns record its identifier, material type, collection date, and mass in grams.

Semistructured

Have fields identified by names or tags, but do not require every record to have the same fixed set of fields. They can include lists and groups of information within a record. JSON and XML can represent this flexible organization. [1] [9]

Example: a JSON document contains sample A01, with a mass of 2 g, and A02, with a mass of 4 g and a list of analyses. Each analysis for A02 includes a date, method, and result. A schema — a set of rules — can define required and optional fields and value types. The file format alone does not determine the classification.

Unstructured

The main content is not organized into predefined analysis fields, such as table columns. [1]

Example: a leaf photograph, an interview recording, or an article’s text. A spot count is obtained through image analysis; record the method and retain its link to the original photograph.

Objects in rows, attributes in columns

In this example, each row represents an object — here, a sample. Each column after the identifier represents an attribute: a recorded characteristic. There are n = 3 objects and d = 5 attributes: a 3 × 5 matrix X. The identifier distinguishes objects and, in this example, is kept outside X.

On narrow screens, scroll tables sideways to see all columns. With a keyboard, focus the table region and use the horizontal arrow keys.

Teaching example: 3 objects × 5 attributes, with fictional values.
Object (ID)MaterialReview priorityAir temperature at collection (°C)Dry mass (g)Linked records
A01leaflow1023
A02rootmedium2046
A03inflorescencehigh3069

Reading row A01: the sample is a leaf with a dry mass of 2 g; 10 °C is the air temperature at collection. Material is categorical; mass and temperature are numerical. For algorithms that require numerical inputs, categories must be encoded.

What values can each attribute take?

A data dictionary specifies meanings, types, units, and permitted values. Qualitative or categorical attributes describe classes. Quantitative or numerical attributes express quantities: discrete values arise from counts; continuous values represent measurements such as mass, size, and distance, recorded with resolution and uncertainty determined by the method. Not every number is a quantity: identification codes are labels.

Dictionary of the table’s five attributes: type, permitted values, and scale.
AttributeTypeValues: examples and interpretationScale
MaterialQualitativeCategories: leaf, root, or inflorescence. Choose one category for each sample.Nominal
Review priorityQualitativeOrdered levels: low < medium < high. The order indicates priority, not equal differences between levels.Ordinal
Air temperature at collectionQuantitative, continuousEnvironmental condition at collection, measured in °C, such as −2; 10; 20.5. Record instrument resolution and measurement uncertainty, when available. Zero is a scale reference, not the absence of temperature.Interval
Dry massQuantitative, continuousA nonnegative measurement in g, such as 2 or 2.5, according to the drying and weighing procedure. Report balance resolution; decimal places alone do not demonstrate precision.Ratio
Linked recordsQuantitative, discreteAn integer count: 0, 1, 2… Fractions are not valid. Zero means no linked records; an unknown count must be marked as missing.Ratio

Missing observations require their own convention, documented separately from valid values. Encoding low, medium, and high as 1, 2, and 3 does not demonstrate equal differences between categories.

Precision, resolution, and uncertainty: precision describes agreement between repeated measurements under specified conditions; resolution is the smallest change that produces a perceptible change in an instrument’s indication. Uncertainty characterizes the dispersion of values that could be attributed to the measured quantity, based on available information. None of these properties is determined by decimal places alone. VIM — International Vocabulary of Metrology. [21]

Practical application: real data and their origins

The teaching table above explains data organization. The cases below apply the same reading to records from the Kannabium database, chemical sources such as ChEBI (a database of chemical entities), and a dataset used on the website. Each row presents a distinct case, not a sample from the same experiment. These cases do not form a matrix of mutually comparable objects.

Object, attribute, and value in four real applications. Data accessed on 18 October 2026.
ObjectAttributeValueType and interpretationOrigin
THCA synthase — structure 3VTECrystallographic resolution2.75 ÅNumerical: resolution of the diffraction data, in ångströms; not the uncertainty of each atom’s position.Kannabium database; structure record 1, checked against RCSB PDB. [6]
Cannabidiol — CBDMolecular formulaC21H30O2Symbolic: describes molecular composition, but does not by itself specify structure or isomer. It is not a quantity to add.ChEBI:69478, an official chemical-data source. [18]
Linalool — monoterpenoidNumber of carbon atoms10Discrete numerical: a count derived from C10 in the formula C10H18O.ChEBI:17580; derived from the formula, not from a sample measurement. [18]
2026 careers sampleNumber of records35Discrete numerical: a count of records with distinct identifiers.Website dataset, version 2.4.0. It does not represent the total number of jobs in the market. [19]

Variable types and measurement scales

A variable is a recorded characteristic. In the sample table, each column represents a variable: material is categorical; mass is numerical. A table can contain variables with different measurement scales.

At Kannabium, Plantae–Animalia connects plant content with studies in animals. THCA synthase is an enzyme in the biosynthesis of tetrahydrocannabinolic acid (THCA). Δ⁹-Tetrahydrocannabinol (THC) is a cannabinoid; cannabinoid receptor type 1 (CB1) is a receptor protein, not a cannabinoid.

Four measurement scales, with definitions and illustrative examples in Kannabium’s Plantae–Animalia context.
GroupScaleDefinitionExample
CategoricalNominal

Categories identified by names, with no order of magnitude. They allow records to be grouped and counted, not levels to be compared.

Relations: x = y or x ≠ y compare labels. Ordering or adding these labels has no quantitative meaning.

Properties: in a single-category field, categories must be mutually exclusive. Frequency is the count for each category; proportion is its share of the total; the mode is the most frequent category, with ties possible. [1] [3]

Molecular identity within the dataset: THCA synthase, Δ⁹-tetrahydrocannabinol (THC), or CB1. Each label identifies the entity described by a record, without ordering them. Functions such as enzyme or receptor should be recorded in another attribute; they do not necessarily form mutually exclusive classes.

Notation: if records A and C have the label THC and B has the label CB1, x_A = x_C and x_A ≠ x_B. Equality refers to labels, not physical molecules. THC and CB1 have no order in this variable.

CategoricalOrdinal

Categories with a defined order. They indicate levels of a characteristic, but order alone does not specify differences or proportions between levels.

Relations: equality and order, such as low < medium < high. The difference high − medium does not provide a quantity.

Properties: codes 1, 2, and 3 can record order, but their numerical intervals do not demonstrate equal distances between responses. The classification allows positions to be compared, not levels to be doubled or halved. [1] [3]

In an illustrative cell-assay example involving the CB1 receptor, responses are grouped as low, medium, or high according to ranges defined for that assay. “High” does not necessarily mean twice “medium”.

Notation: high > medium > low indicates only the order of response classes. If an original assay measurement is available, it should be recorded with its unit, separately from the category.

NumericalInterval

Equal differences represent equal changes. The starting point is a reference, not absence of what is measured.

Relations: equality, order, and differences, such as Δ = x − y. Equal differences have the same meaning in the same unit. The ratio x / y between original values depends on the chosen origin and does not express a proportion of the quantity.

Properties: zero is conventional. The scale indicates which comparisons are meaningful; it does not guarantee measurement quality or eliminate errors. [1] [3] [4]

Illustrative dates of cannabinoid analyses in a Cannabis sample: June 1, 3, and 5, 2026. Two days elapsed between the first and second analyses, and also between the second and third. The dates indicate when the analyses occurred; monitoring duration is calculated from the difference between dates.

Notation: let d₁, d₂, and d₃ denote these dates: d₂ − d₁ = d₃ − d₂ = 2 days. Dividing dates does not indicate how many times later an analysis occurred. Calculated duration is a different variable, with zero for no elapsed time.

NumericalRatio

Measurements or counts in which zero means absence of the recorded quantity. Differences and proportions, such as twice and half as much, can be compared.

Relations: equality, order, differences, and ratios. For values of the same quantity in the same unit, x / y compares magnitudes, provided y ≠ 0.

Properties: zero means absence of the defined quantity. Sums, products, and other operations require meaningful interpretation and consistent units; the scale does not authorize arbitrary calculations. Ratio scale does not mean that values must be rational numbers written as fractions p/q. [1] [2] [3]

Number of THC molecules in two illustrative models: A contains 1; B contains 2. B has twice as many of these molecules. A model with 0 THC molecules does not contain this compound. If the count was not performed, the value is missing, not zero.

Notation: n_A = 1, n_B = 2, and n_B / n_A = 2; therefore, n_B = 2 × n_A. n = 0 records no THC included in the model; it does not prove absence of the compound in a physical sample.

Temperature example: 10, 20, and 30 °C have equal intervals of 10 °C. However, 20 °C is not twice the thermodynamic temperature of 10 °C: Celsius uses a conventional zero, not the absence of temperature. Compare differences in °C, not ratios between these values. [3]

Response ranges depend on the assay and its controls; they are not a universal classification. Counting molecules in a model does not measure binding affinity or biological effect.

Codes such as 001 and 002 identify samples. Although they use numbers, they do not represent quantities or indicate whether one sample is larger or smaller than another.

Sequences, structures, and other biological data

In bioinformatics, sequence, structure, location, and image records represent different aspects of the objects being studied. Identifying what each record describes is the first step in choosing how to analyze it.

Biomolecular structures
The PDB holds three-dimensional models of proteins, nucleic acids, and their complexes. A sequence records the order of molecular units; a model records their spatial arrangement. In PDBx/mmCIF (a structural-data format), modeled atom positions are recorded by x, y, and z coordinates in ångströms (Å), a unit of length used at the atomic scale. Regions of the molecule may lack coordinates. Real example: the THCA-synthase model obtained by X-ray crystallography, PDB 3VTE. [6] [7]
Geospatial data
Link information to a location. Example: a botanical-record table with species, latitude, longitude, date, and source. Coordinates allow the records to be plotted on a map. When combining records, check the coordinate reference system and location uncertainty, where reported. The map shows available records; an area without records does not establish that the species is absent. [8]
Images, audio, and video
Record visual or sound information. Examples: a trichome photograph, an interview recording, and a molecular-simulation video. The video shows the molecule as an image; calculating distances between atoms requires the structure or trajectory coordinates. A trajectory records atom positions over time. [1]

Data representations and multimodal combinations

The list brings together ways to represent data, not mutually exclusive classes: a sequence may be recorded as text, and a video may include audio. Multimodal means combining modalities, not an additional basic modality. Structure and measurement scale are separate criteria.

Representations, combinations, and application examples.
Representation or combinationExample and what to retain
AudioInterview recording: consent, sampling rate, and transcript provenance.
Multimodal combinationLeaf image with spectrum and sample table: identifiers that connect the same specimen across modalities.
GraphsMolecular interaction network: node identities and the evidence behind each connection.
ImagesTrichome photograph: resolution, scale, and acquisition conditions.
Biological sequencesDNA, RNA, or proteins: identifier, organism, alphabet, and record version.
Time seriesRepeated measurements from one sensor: timestamps, sampling interval, and gaps.
TextScientific article: authorship, page or section, and reuse conditions.
VideoPlant growth recording: frame rate, duration, and acquisition conditions.

Biological sequences: DNA, RNA, and proteins

Sequences represent an order: of nucleotides, the units that make up DNA and RNA, or of amino acids in proteins. After processing and annotation, they may become part of specialized databases: GenBank holds nucleotide sequences; UniProt holds protein sequences and information about their function. [5] [17]

For analysis, use a defined sequence version to fix the content being studied. Annotations may be updated without a sequence change; a sequence version does not freeze the entire database entry. Retain the identifier, version, and access date. [5]

In FASTA, each record begins with a header followed by the sequence. The format’s rules organize this information in a text file. The term “raw” describes the data’s processing stage; structure describes how the content is organized. Explore these elements in Read a real FASTA from the project. [5]

From the original record to a dataset

Usable data require more than a valid format. Taxonomic identification, the origin of each observation, and the meaning of each measurement must remain traceable.

Taxonomic identity
Keep the name as published, its authorship, and the voucher specimen when available. Record the normalized name, reference database, and access date or version separately. An automated name match does not confirm plant identification. [10]
Traits in GBIF text
Katori and Oguro recovered traits from descriptions associated with GBIF records. When turning text into analytical fields, preserve the original passage, record identifier, extraction rule, and unit. Increased coverage does not guarantee comparable measurements or inclusion of a particular species. [11]
Value origin and transformation
Indicate whether a value was observed or measured, derived by calculation, or imputed to fill a gap or replace an invalid record. Preserve the original value and document the transformation, method, and uncertainty. An imputed estimate must not be presented as an observed measurement. Statistical dependence between records is a separate issue. [22]
Organ, method, and detection
Jin and colleagues profiled inflorescences, leaves, stem barks, and roots separately. Preserve organ, material, unit, and replication. Not detected, not measured, and a concentration of zero are different situations. Non-detection depends on the method and detection limit; it does not demonstrate absence of the compound. [12]

Applications from the literature: Ferreira de Lima and colleagues (2026), Katori and Oguro (2026), and Jin and colleagues (2020). These are external studies, not experiments conducted by Kannabium.

Before analysis, record available sources, identifiers, versions, and dates. Check field meanings, units, format, and how missing values are recorded.

To compare masses, check the unit, material type, and condition, such as fresh or dry. An empty field does not mean a mass of zero; consult the documentation before handling missing values.

The 3 × 5 table uses fictional values. Real cases have identified sources and include external records and derived values; their presence in this guide does not make them experimental results produced by Kannabium.

Continue to the data, models, and governance guide →

Further reading on data analytics and decision support: Sharda, Delen, and Turban. [13]

Sources and further reading

Bracketed numbers point to the references for each topic. Books [13]–[15] are further reading on data analysis, machine learning, and decision support. Teaching examples connect these concepts with the Kannabium context.

↑ Back to top