From Complex to Simple and Back: Mathematical Abstraction of Life's Logic Reveals Disease-Driving Essentials

1 | Page From Complex to Simple and Back: Mathematical Abstraction of Life’s Logic Reveals Disease-Driving Essentials Running title: Life computes in logic— biology and medicine should too Authors Saptarshi Sinha1* and Pradipta Ghosh1,2* Departments of 1Cellular and Molecular Medicine and 2Medicine, University of California, San Diego, CA, 92093, USA. *Correspondence to [email protected] (SS) or [email protected] (P.G). Keywords (two to six) Boolean Implication Networks · Invariants · Systems Biology · StepMiner · Dose–Response Alignment · Cellular Intelligence · Function-Agnostic Modeling Abstract (125 words) Biology and medicine mistake complexity for understanding. We build black-box AI models and stockpile terabytes of omics data, yet the logic of life remains hidden in plain sight. The cell, nature’s smallest decision-maker computes not by probability, but by logic: “on” or “off,” “commit” or “retract.” Boolean mathematics decodes this digital decision-making, transforming analog molecular noise into invariant “if–then” rules that persist across tissues, species, and diseases. Simultaneously, by quantifying how populations shift between binary states, Boolean frameworks map disease as an analog continuum—dynamic, graded, reversible and objectively measurable. By deciphering both the binary decisions and the graded population shifts that shape disease, this approach replaces static biomarkers with mechanistic rules—and advances a new premise: if life computes in logic, medicine should too. 2 | Page Glossary: 1 StepMiner: A threshold-detection algorithm that converts noisy continuous gene expression into binary high 2 (1) /low (0) states by identifying the most significant step change in expression. 3 4 Invariant (Logical Constraint): A molecular relationship that consistently holds under diverse conditions, 5 revealing regulatory constraints essential to biological stability. 6 7 Boolean Implication Relationships (BIRs): Is a pair-wise gene expression relationship between two genes 8 with respect to their gene expression values that define directional “if-then” relationships between genes in 9 binary high/low expression states, revealing invariant molecular rules. 10 11 Boolean Implication Networks (BINs): A directed network of binary “if–then” relationships between genes 12 (e.g., A high ⇒ B low) that reflect invariant molecular dependencies across samples, tissues, or species. 13 14 Threshold (Binary Decision Point): The inflection point at which a continuous biological signal flips into a 15 decisive action — “on/off,” “commit/retract,” “live/die.” 16 17 Boolean Trajectory (Continuum): A directional path through gene-state space (e.g., A low ⇒ B high ⇒ C 18 high) that maps continuum states of progression from health toward disease — and potentially back. 19 20 Continuum: a continuous sequence in which adjacent elements are not perceptibly different from each other, 21 although the extremes are quite distinct. 22 23 Dose–Response Alignment (DoRA): A systems biology principle in which feedback circuits ensure 24 proportionality between molecular input and output, preserving information fidelity. 25 26 Function-Agnostic Benchmarking: Model comparison based on raw gene-state logic rather than evolving 27 pathway annotations, enabling reproducible and unbiased assessment of disease fidelity. 28 29 Digital Biomarker: A disease-relevant binary or composite gene-state signature that signals a shift in core 30 biological logic before clinical manifestation. 31 32 Systems Biology: An interdisciplinary field that studies biological systems as networks of interacting 33 components, seeking emergent principles that govern cellular behavior across scales. 34 35 3 | Page Stochastic: Referring to processes influenced by randomness or noise. Biological systems often exhibit 36 stochasticity at the molecular level, yet produce robust, reproducible outcomes at the cellular level. 37 38 Probabilistic Models: Mathematical frameworks that describe outcomes as likelihoods rather than certainties. 39 Common in gene-network inference but limited in revealing directional or causal logic. 40 41 Analog vs. Digital Biology: Analog systems represent continuous variation (e.g., graded gene expression), 42 while digital systems rely on discrete states (e.g., on/off). The Boolean framework translates biology’s analog 43 complexity into digital clarity. 44 45 Ontology: A curated hierarchical system (e.g., GO, Reactome) used to assign functional meaning to genes — 46 invaluable but incomplete and constantly changing. 47 48 System-Level Majority Rule: A network principle where collective shifts in cell-state logic reflect coordinated 49 population behavior, even when evaluated at gene-level resolution. 50 51 Causal Constraint: A directional logical rule that defines what a biological system must do (or cannot do) 52 when a specific molecular state occurs. 53 54 Artificial Intelligence (AI): Computational systems capable of learning patterns or behaviors from data. In 55 biology, AI models simulate complexity but often lack interpretability. 56 57 Large Language Models (LLMs): A class of AI models trained on vast textual data to predict and generate 58 language. LLMs excel at prediction and summarization but do not inherently understand biological causality 59 60 Machine learning: A branch of artificial intelligence (AI) and computer science which focuses on the use of 61 data and algorithms to find patterns that humans cannot find. 62 63 KEY FIGURE: Figure 1: Life’s Logic, Abstracted Mathematically as Five Rules64 4 | Page When Biology Defies Mathematics The laws of nature, from Newton’s gravitation to Einstein’s relativity, astonish us by their simplicity, invariance, and predictive power. These mathematical regularities, i.e., conditional statements describing what must follow from what is, form the scaffolds of physics. Biology (and Medicine), however, refuses such precision. For centuries, the living world has resisted the kind of mathematical formalism that so elegantly explains the cosmos. The cell—life’s smallest decision-making unit—is a system of staggering complexity, yet it operates on remarkably simple, universal principles. Within it, thousands of molecules interact through circuits that sense, decide, and act. These decisions—whether to divide, differentiate, migrate, or die—trace the continuum of transition states between health and disease. Despite breathtaking advances in molecular profiling and AI, our ability to extract the invariant rules that govern these transitions remains limited. Some biologists argue this is not due to ignorance, but due to impossibility2: states of life—evolution, emergence, adaptation—defies closure within fixed mathematical systems. Concepts such as “genes”, “species”, or “fitness” remain context dependent and fluid. Biology transcends computation: it is a science of thresholds, feedback, and exceptions. And yet, if mathematics cannot define life, perhaps it can describe its logic. Boolean frameworks, through sophisticated thresholding3, embrace rather than erase biological complexity, translating noise into rules, randomness into predictability. Instead of exact equations, they seek invariant rules—binary “if–then” relationships that persist across tissues, species, and diseases. These Boolean implications, first formalized by Sahoo et al.4,5, expose the simple decision-making architecture within the cellular storm. At its core lies StepMiner1, an thresholding algorithm that converts noisy gene-expression data into binary outcomes—“low (0)” or “high (1)”—mirroring how cells themselves decide to act. In doing so, Boolean logic offers something rare in modern biology: a mathematical abstraction that captures life’s decision logic, not its descriptive chaos. In an era of black box AI, Boolean framework stands apart. It is not a simulation of complexity but a simplification that reveals universals, a quiet order beneath the stochastic noise of biology, governed by rules as elegant and predictive as any law of physics [Figure 1; BOX 1]: 5 | Page Rule 1 | Life Computes in Thresholds Every living cell is a decision engine. Molecules fluctuate continuously, but at critical junctures the system commits — to divide or arrest, differentiate or retain stemness, activate immunity or remain quiescent. These commitments are neither probabilistic nor gradual; they are threshold crossings. A cell does not partly divide, somewhat differentiate, or tentatively die. It flips. Thresholds are evolution’s answer to noise: regulatory circuits convert variable molecular inputs into binary outcomes, i.e., the molecular equivalents of yes/no, go/stop, live/die. These switches mark the points where possibility becomes physiological action. The Boolean framework captures this architecture mathematically. At its core lies StepMiner1, an adaptive thresholding algorithm that detects inflection points in gene-expression profiles (Figure 2A). StepMiner fits a step function to each gene’s sorted expression values to determine the most significant transition separating “off (0)” from “on (1)” states. This converts noisy continuous data into clear low (0) or high (1) states, with a noise buffer removing uncertainty around the threshold. This digital abstraction is not an oversimplification; it mirrors how cells compute: • Fate choice: mutually inhibitory transcription factors enforce lineage commitment. • Apoptosis: feedback-driven switches distinguish survival vs. death. • Immune activation: T cells fully engage only beyond receptor–signal thresholds. By imposing such thresholds, the Boolean framework transforms high-dimensional, stochastic omics data into interpretable decision landscapes, and converts continuous molecular inputs into binary outcomes through signal transduction, gene regulation, and metabolic feedback. Box 1. The Five Rules of Biological Logic 1. Life Computes in Thresholds – Thresholding algorithms such as StepMiner1 convert noisy expression into decisive 0/1 states, revealing where cells commit to differentiation, survival, activation, or death — the binary logic underlying biological decisions. 2. Noise Hides Invariants – Boolean implication reveals relationships that persist despite biological or experimental noise, exposing rules that hold across tissues, species, and disease. 3. Feedback Is Fidelity – Boolean edges trace push–pull signaling architectures that maintain Dose–Response Alignment (DoRA), identifying the checkpoints where proportionality is preserved when biology maintains order and where dysregulation begins. 4. Disease Is a Reversible Continuum – Boolean implications define the directionality of health-todisease transitions, while analog composite scores (StepMiner-normalized) position each sample along a reversible, quantifiable continuum. 5. Function Follows Logic, Not Annotation – Composite scores from data-derived Boolean networks retain analog behavior while exposing causal constraints and enable function-agnostic benchmarking, defining biological meaning through empirical invariants rather than evolving ontologies, thereby improving target discovery and model fidelity. 6 | Page Crucially, digital abstraction at this step does not discard analog information that is vital to track cellular processes in multicellular life that continues to evolve through complex cell-cell and cell-environment crosstalk. The StepMiner-normalized continuous score is retained and becomes foundational in: Rule 4: placing samples along the analog health → disease continuum and in Rule 5: quantifying population-level drift and therapeutic reversibility. From Analog Chaos to Digital Clarity: StepMiner and the Birth of BINs Traditional gene-network methods, e.g., co-expression, mutual information, Bayesian inference, are symmetric and probabilistic, capturing correlation but not direction. They struggle to extract meaning or retain reproducibility in the presence of real-world heterogeneity and noise. Boolean Implication Networks (BINs)3 changed this paradigm by introducing logic as biology’s native language: simple “if–then” rules linking high/low gene-expression states. The resulting network of Boolean implication relationships (BIRs)—such as “A high ⇒ B high” or “A high ⇒ B low” (Figure 2B)—forms a digital fingerprint of cellular logic. Each gene becomes a logical variable, and each Boolean implication forms a constraint — a rule biology consistently obeys. Together, these relationships produce a digital fingerprint of how the system thinks, revealing direction, dependency, and hierarchy as statistical facts rather than assumptions. In biology and in math, robustness arises from thresholds. Life decides digitally and the Rule 1 in the Boolean framework measures those decisions. Rule 2 | Noise Hides Invariants If thresholding exposes the moment of decision, Boolean implication uncovers the rules that endure (Figure 2B). Biology is inherently noisy—gene expression fluctuates, proteins misfold or misfire, pathways crosstalk— yet within this apparent chaos lie invariants: relationships that remain true across tissues, species, and perturbations. These are nature’s constraints, the hidden grammar of gene-regulatory logic that evolution preserves. The Boolean framework detects these invariants not just through symmetric correlation but also through asymmetric logic (Figure 2B). For every gene pair (A, B) across thousands of samples, their joint expression pattern, after thresholding, divides the samples into four groups (or four quadrants in a two-dimensional plot)— A low/high × B low/high. When one quadrant is significantly underpopulated, indicating that a particular combination of states almost never occurs, an if–then Boolean implication is established (Figure 2B). This asymmetric exclusion reveals the directional constraints that biology consistently enforces. 7 | Page The Six Boolean Relationships as Natural Invariants 1. Four asymmetric: “A high ⇒ B high,” “A high ⇒ B low,” “A low ⇒ B high,” “A low ⇒ B low” 2. Two symmetric: “Equivalent” (A ⇔ B), or “Opposite” (A high ⇒ B low AND B high ⇒ A low) Each relationship represents an empirical invariant, a directional, data-driven relationship that holds across diverse biological contexts. These are not hypotheses; they are statistical facts, distilled directly from huge real-world expression data. They define which molecular states co-exist, which exclude one another, and which form nested hierarchies of regulation. They represent constraints enforced at the cellular level and sustained through heterotypic feedback and crosstalk among cells within tissues—emergent rules that hold at the population scale. Across cancers, developmental programs, and cross-species comparisons, such invariants repeatedly surface: • B-cell maturation: Stem marker high ⇒ differentiation marker low pinpoints transitional genes defining the midpoints of developmental hierarchies6. These Boolean intermediates helped identify previously unrecognized regulators of lineage progression—genes silent in stem states but active as differentiation begins. • Bladder cancer: KRT5 high ⇒ KRT20 low demarcates basal progenitors from luminal cells7, defining a binary architecture of epithelial plasticity. This rule persists across patient samples and species, signifying a deeply conserved regulatory toggle between basal identity and terminal differentiation. • Colon tumors: CA1 high ⇒ KRT20 high captures a nested lineage constraint in differentiated tumors, reflecting hierarchical fidelity within intestinal epithelium. Conversely, ALCAM or CCDC88A high ⇒ CDX2 or PRKAB1 low marks a stemness axis—cells reverting toward an undifferentiated, regenerative program, often predictive of therapeutic resistance and poor prognosis. • Lung injury and fibrosis: ACE2 high ⇔ IL15/IL15RA high signals a conserved epithelial–immune coregulation circuit stable across human and mouse models8-11. • Macrophage polarization: Conserved implication patterns trace immune trajectories from reactivity to tolerance, revealing logical paths that remain fixed across tissues and species12. • Inflammatory bowel disease (IBD): Genes preserving epithelial integrity and energy homeostasis share a high ⇒ low relationship with those driving inflammation and fibrosis—computationally trackable across organoids, animal models, and patient biopsies.3,13. Together, these and numerous other examples14-16 demonstrate how the Boolean framework reproducibly extracts invariant logic beneath biological diversity. Whether in hematopoiesis or fibrosis, immune tolerance or epithelial regeneration, the same mathematical grammar applies, i.e., directional relationships, conserved across scales, that define what biology allows or forbids. Boolean relationships thus emerge as constraints imposed by life itself; rules so fundamental that they persist through evolution and pathology alike. The Boolean 8 | Page logic redefines “mechanism” not as a static pathway diagram but as a set of logical constraints which represent the boundaries of biological possibility that remain stable through evolution, adaptation and disease. Rule 3 | Feedback Is Fidelity Life maintains order not by silencing noise but by aligning feedback to preserve proportionality. This is better known as Dose–Response Alignment17 (DoRA), a principle that describes how cells preserve faithful information transfer by aligning the input-to-output relationship across signaling cascades18. When feedback loops achieve DoRA, downstream responses scale predictably with receptor activation, converting molecular chaos into coherent behavior. Mechanistically, DoRA is achieved not by fine-tuning every parameter but by architectural motifs19,20—push-pull regulatory networks in which the active form of a signaling species drives output (“push”) while the inactive form counter-acts it (“pull”)21. The Boolean framework captures these equilibria at the transcriptomic level in complex biological circuits20,22. Each Boolean edge, A high ⇒ B high or A high ⇒ B low, represents a feedback-stabilized dependency, the digital trace of push–pull motifs that maintain biological fidelity. By elevating DoRA from molecular cascades to transcriptomic logic, the Boolean framework reveals the transistor-like decision points in cellular regulation— where biology imposes invariant logic rather than variable curves. These are the nodes where “noise” is filtered, fidelity is enforced, and decisions become digital. In this way, the DoRA concept serves as the bridge from analog chaos of molecular networks to the digital clarity of Boolean implication. Remarkably, Boolean relationships often align with protein-level behaviors (with surprising fidelity), as confirmed by cytochemistry-based validations3,23-25. This suggests that many transcriptional implications are reinforced by feed-forward loops26 coupling transcription and translation. Still, not all alignments signify activation: negative feedback loops27 may sustain transcriptional signals while functionally driving resolution or repair. In disease contexts, the Boolean framework3 helps disentangle these dualities—revealing when persistence of expression reflects stability, and when it encodes homeostatic correction. Ultimately, feedback is fidelity: Boolean invariants reveal where biology enforces proportionality, transforming stochastic expression into reproducible logic. Rule 4 | Disease Is a Reversible Continuum, Not an Absolute Category Disease rarely begins at diagnosis. Across cancers, fibrosis, autoimmunity, and degeneration, decades of evidence show that pathology emerges as a slow, often reversible drift, either genetic, epigenetic, or posttranslational state-dependent, long before clinical diagnostic thresholds are crossed28-33. 9 | Page The Boolean framework3 captures this continuity by mapping BIRs that that define when regulatory decisions flip from stability to instability (Figure 3A-C). Yet the framework also preserves analog information: StepMinernormalized composite scores (modified Z-scores; elaborated in Rule 5) quantify how many cells occupy each state. This enables samples to be positioned along a disease trajectory, revealing graded physiological change. In other words: BIRs define direction, i.e., the order in which transitions occur, and composite scores define position, i.e., how far a sample has progressed. Using graph-based temporal networks, the framework extracts pseudotime and state-transition paths from crosssectional datasets (Figure 3D-G), anchoring each patient, organoid, or perturbation along a quantifiable, twomode health–disease landscape: • digital: “which switches have flipped?” • analog: “how far has the switch propagated across the system?” A population-level majority-rule principle34 ensures that Boolean decisions scale to tissue behavior: when most cells enter a disease-aligned state, the phenotype follows — as validated in macrophage polarization studies spanning reactive ↔ tolerant extremes12,35. This majority rule is supported by its demonstrated effectiveness in achieving consensus and coordinating behavior in complex, interconnected networks, especially when information transfer is decentralized, noisy or incomplete. When most cells shift into a disease-associated state, the tissue-level phenotype follows. Thus, the framework provides testable insight into where recovery remains possible and which interventions may revert state drift, already demonstrated in multiple biological contexts3,12,13,23,36,37. Disease, then, is not an endpoint but a reversible trajectory within a dynamic Boolean landscape, one that can be stabilized, experimentally perturbed, or therapeutically restored. It is noteworthy that the resulting continuum map is data-dependent and can be tuned to clinically meaningful endpoints. in inflammatory bowel disease (IBD), Boolean trajectories of disease processes3 were refined to reflect the most stringent criteria for regulatory approval, i.e., endoscopic and histologic remission13. Thus, Rule 4 elevates disease modeling from descriptive categorization to predictive, clinically actionable guidance. In essence, where conventional models categorize, the Boolean framework quantifies, turning disease into a reversible topography of logic and degree. Rule 5 | Function Follows Logic, Not Annotation Only ~8.2% of the human genome is under functional constraint, and just ~1.5% encodes proteins (estimated to be in the range of 19,587–20,24538-40) responsible for all cellular activity41. Even with structural breakthroughs 16 | Page Outstanding Questions 35 • Can Boolean logic predict therapeutic reversibility? 36 If disease progression follows directional Boolean trajectories, can reversing those edges restore normal 37 decision logic — predicting therapeutic success before clinical response is visible? How early can Boolean 38 drift be detected, and can intervention redirect disease prior to irreversible pathology? 39 • Is feedback fidelity the earliest sign of disease? 40 If health is maintained through aligned input–output relationships (DoRA), does pathology first emerge as a 41 measurable breakdown in feedback fidelity — driven by environmental exposures and aging rather than 42 mutation alone? Could DoRA collapse serve as a universal early-warning biomarker? 43 • What defines the point of no return? 44 Boolean transitions can flag pre-symptomatic changes, but when does a reversible drift become a locked-45 in state? Can Boolean rules map—and therapeutically reopen—the narrowing window for recovery along 46 the disease continuum? 47 • Could Boolean logic serve as the foundational model for explainable biological intelligence? 48 With transparent causal structure, can Boolean rules train LLMs and virtual cells to reason like biology — 49 discrete decisions with graded population shifts — enabling AI that is not merely predictive, but 50 interpretable and reversible? 51 • Will function-agnostic logic replace ontology in how we benchmark disease models and discover 52 drugs? 53 If function-agnostic, data-derived logic consistently outperforms curated pathways, should biology shift 54 from describing networks to uncovering the rules that govern them?55 17 | Page Acknowledgments 56 The authors wish to apologize for not citing many important publications on this topic due to the limit of 57 references allowed. P.G. is supported by NIH grants R01-AI141630 and R01-AI55696. PG was also 58 supported by the Leona M. and Harry B. Helmsley Charitable Trust and the Propel a Cure Foundation. 59 S.S. was supported through The American Association of Immunologists (AAI) Intersect Fellowship 60 Program for Computational Scientists and Immunologists. 61 62 Declaration of interests 63 The authors declare no conflicts of interest.64 18 | Page Figure Legends: Figure 1. The Boolean Framework — From Analog Chaos to Digital Clarity: A minimalist framework that converts transcriptomic noise into logical maps of cellular decision-making, revealing invariant edges, feedback fidelity, Boolean trajectories linking health and disease, complete suite of gene signatures to track those trajectories. 19 | Page Figure 2. Threshold-Based Boolean Relationships Reveal Universal Constraints in Gene Expression A. Schematic illustrates the application of the StepMiner algorithm to define discrete expression thresholds for Gene A and Gene B, converting continuous microarray data into binary states (high vs. low). These thresholds enable the identification of Boolean relationships between genes. B. A spectrum of directional and symmetric gene expression relationships, categorized as asymmetric (e.g., “A high ⇒ B high”, “C high ⇒ D low”) and symmetric (e.g., equivalence, opposition). Each scatter plot is divided by threshold lines, revealing regions of biological impossibility—highlighted in orange—as “Forbidden possibilities.” These voids reflect universal constraints imposed by cellular logic, suggesting that Boolean relationships are not merely descriptive but mechanistic, conserved across evolution, adaptation, and pathology. 20 | Page Figure 3. Boolean Logic Enables a Scalable Framework for Disease Modeling and Therapeutic Discovery This workflow outlines a data-driven approach to understanding disease progression using gene expression profiles and Boolean logic. A. Gene expression datasets from healthy and diseased individuals serve as the foundation. B. A Clustered Boolean Implication Network (BIN) organizes genes into clusters based on invariant logical relationships. C. Boolean edges—such as equivalence, directional implications, and oppositions—define stable constraints between clusters. D. These constraints enable construction of a continuum model of cellular states, tracing trajectories from health to disease. E. Boolean paths reveal nested hierarchies and sequential transitions in gene expression, offering mechanistic insight. F-G. Machine learning models (F) interpret these paths, in light of sample distribution (healthy vs. disease) within the BIN, enhancing interpretation and explanatory power. The resulting health-disease continuum map (G) visualizes biological transitions and tipping points. H. Composite gene signatures derived from BINs are function-agnostic yet biologically grounded. I. Applications include cohort validation, objective disease severity scoring, benchmarking disease models, predictive modeling of clinical endpoints, iterative model refinement, and therapeutic target identification. Together, these steps demonstrate how Boolean logic transforms gene expression data 21 | Page Figure 4. From Annotation to Logic: How The Boolean Framework Defines Function from Data Left (Annotation): Traditional biology depends on human-defined ontologies (e.g., Gene Ontology, Reactome, KEGG), where function is imposed top-down through evolving and incomplete categorical hierarchies. Middle (Abstraction): Modern machine-learning frameworks simplify complexity into latent dimensions, compressing continuous expression data into lower-dimensional abstractions, but often at the cost of interpretability. Right (Logic): The Boolean framework transcends both by deriving function directly from data. It binarizes expression profiles into “low/high” states, discovers invariant Boolean “if–then” relationships, and reconstructs a transparent decision network that captures the causal logic of biological systems. 22 | Page REFERENCES: 1 Sahoo, D., Dill, D. L., Tibshirani, R. & Plevritis, S. K. Extracting binary signals from microarray timecourse data. Nucleic Acids Res 35, 3705-3712 (2007). https://doi.org/10.1093/nar/gkm284 2 Garte, S., Marshall, P. & Kauffman, S. The Reasonable Ineffectiveness of Mathematics in the Biological Sciences. Entropy (Basel) 27 (2025). https://doi.org/10.3390/e27030280 3 Sahoo, D. et al. Artificial intelligence guided discovery of a barrier-protective therapy in inflammatory bowel disease. Nat Commun 12, 4246 (2021). https://doi.org/10.1038/s41467-021-24470-5 4 Sahoo, D. The power of boolean implication networks. Front Physiol 3, 276 (2012). https://doi.org/10.3389/fphys.2012.00276 5 Sahoo, D., Dill, D. L., Gentles, A. J., Tibshirani, R. & Plevritis, S. K. Boolean implication networks derived from large scale, whole genome microarray datasets. Genome Biol 9, R157 (2008). https://doi.org/10.1186/gb-2008-9-10-r157 6 Sahoo, D. et al. MiDReG: a method of mining developmentally regulated genes using Boolean implications. Proc Natl Acad Sci U S A 107, 5732-5737 (2010). https://doi.org/10.1073/pnas.0913635107 7 Volkmer, J. P. et al. Three differentiation states risk-stratify bladder cancer into distinct subtypes. Proc Natl Acad Sci U S A 109, 2078-2083 (2012). https://doi.org/10.1073/pnas.1120605109 8 Sinha, S. et al. COVID-19 lung disease shares driver AT2 cytopathic features with Idiopathic pulmonary fibrosis. EBioMedicine 82, 104185 (2022). https://doi.org/10.1016/j.ebiom.2022.104185 9 Sahoo, D. et al. AI-guided discovery of the invariant host response to viral pandemics. EBioMedicine 68, 103390 (2021). https://doi.org/10.1016/j.ebiom.2021.103390 10 David, P. et al. MDA5-autoimmunity and interstitial pneumonitis contemporaneous with the COVID-19 pandemic (MIP-C). EBioMedicine 104, 105136 (2024). https://doi.org/10.1016/j.ebiom.2024.105136 11 Ghosh, P. et al. An Artificial Intelligence-guided signature reveals the shared host immune response in MIS-C and Kawasaki disease. Nat Commun 13, 2687 (2022). https://doi.org/10.1038/s41467-02230357-w 12 Ghosh, P. et al. Machine learning identifies signatures of macrophage reactivity and tolerance that predict disease outcomes. EBioMedicine 94, 104719 (2023). https://doi.org/10.1016/j.ebiom.2023.104719 13 Sinha, S. et al. F.O.R.W.A.R.D: A Data-Driven Framework for Network-Based Target Prioritization in Drug Discovery. bioRxiv (2025). https://doi.org/10.1101/2024.07.16.602603 14 Zage, P. E. et al. Identification of a novel gene signature for neuroblastoma differentiation using a Boolean implication network. Genes Chromosomes Cancer 62, 313-331 (2023). https://doi.org/10.1002/gcc.23124 15 Vo, D., Ghosh, P. & Sahoo, D. Artificial intelligence-guided discovery of gastric cancer continuum. Gastric Cancer 26, 286-297 (2023). https://doi.org/10.1007/s10120-022-01360-3 16 Ghosh, P. et al. AI-assisted discovery of an ethnicity-influenced driver of cell transformation in esophageal and gastroesophageal junction adenocarcinomas. JCI Insight 7 (2022). https://doi.org/10.1172/jci.insight.161334 23 | Page 17 Qiao, L., Ghosh, P. & Rangamani, P. Design principles of improving the dose-response alignment in coupled GTPase switches. NPJ Syst Biol Appl 9, 3 (2023). https://doi.org/10.1038/s41540-023-00266-9 18 Nunns, H. & Goentoro, L. Signaling pathways as linear transmitters. Elife 7 (2018). https://doi.org/10.7554/eLife.33617 19 Milo, R. et al. Network motifs: simple building blocks of complex networks. Science 298, 824-827 (2002). https://doi.org/10.1126/science.298.5594.824 20 Alon, U. Network motifs: theory and experimental approaches. Nat Rev Genet 8, 450-461 (2007). https://doi.org/10.1038/nrg2102 21 Yan, L., Ouyang, Q. & Wang, H. Dose-response aligned circuits in signaling systems. PLoS One 7, e34727 (2012). https://doi.org/10.1371/journal.pone.0034727 22 Qiao, L. et al. A circuit for secretion-coupled cellular autonomy in multicellular eukaryotic cells. Mol Syst Biol 19, e11127 (2023). https://doi.org/10.15252/msb.202211127 23 Sinha, S. et al. CANDiT: A machine learning framework for differentiation therapy in colorectal cancer. Cell Rep Med, 102421 (2025). https://doi.org/10.1016/j.xcrm.2025.102421 24 Dalerba, P. et al. CDX2 as a Prognostic Biomarker in Stage II and Stage III Colon Cancer. N Engl J Med 374, 211-222 (2016). https://doi.org/10.1056/NEJMoa1506597 25 Dalerba, P. et al. Single-cell dissection of transcriptional heterogeneity in human colon tumors. Nat Biotechnol 29, 1120-1127 (2011). https://doi.org/10.1038/nbt.2038 26 Mangan, S. & Alon, U. Structure and function of the feed-forward loop network motif. Proc Natl Acad Sci U S A 100, 11980-11985 (2003). https://doi.org/10.1073/pnas.2133841100 27 Krishna, S., Andersson, A. M., Semsey, S. & Sneppen, K. Structure and function of negative feedback loops at the interface of genetic and metabolic networks. Nucleic Acids Res 34, 2455-2462 (2006). https://doi.org/10.1093/nar/gkl140 28 Curtius, K., Wright, N. A. & Graham, T. A. Evolution of Premalignant Disease. Cold Spring Harb Perspect Med 7 (2017). https://doi.org/10.1101/cshperspect.a026542 29 Gerstung, M. et al. The evolutionary history of 2,658 cancers. Nature 578, 122-128 (2020). https://doi.org/10.1038/s41586-019-1907-7 30 Beason-Held, L. L. et al. Changes in brain function occur years before the onset of cognitive impairment. J Neurosci 33, 18008-18014 (2013). https://doi.org/10.1523/JNEUROSCI.1402-13.2013 31 Caselli, R. J. et al. Neuropsychological decline up to 20 years before incident mild cognitive impairment. Alzheimers Dement 16, 512-523 (2020). https://doi.org/10.1016/j.jalz.2019.09.085 32 Vestergaard, M. V., Allin, K. H., Poulsen, G. J., Lee, J. C. & Jess, T. Characterizing the pre-clinical phase of inflammatory bowel disease. Cell Rep Med 4, 101263 (2023). https://doi.org/10.1016/j.xcrm.2023.101263 33 Evans-Molina, C. et al. β Cell dysfunction exists more than 5 years before type 1 diabetes diagnosis. JCI Insight 3 (2018). https://doi.org/10.1172/jci.insight.120877 34 Tamir, R., Livshits, A. & Shadmi, Y. Simple Majority Consensus in Networks with Unreliable Communication. Entropy (Basel) 24 (2022). https://doi.org/10.3390/e24030333 24 | Page 35 Katkar, G. & Ghosh, P. Macrophage states: there's a method in the madness. Trends Immunol 44, 954964 (2023). https://doi.org/10.1016/j.it.2023.10.006 36 Sinha, S. et al. Growth signaling autonomy in circulating tumor cells aids metastatic seeding. PNAS Nexus 3, pgae014 (2024). https://doi.org/10.1093/pnasnexus/pgae014 37 Katkar, G. D. et al. Distinct Colitis-Associated Macrophages Drive NOD2-Dependent Bacterial Sensing and Gut Homeostasis. bioRxiv (2025). https://doi.org/10.1101/2025.01.21.634180 38 Gaudet, P. et al. The neXtProt knowledgebase on human proteins: 2017 update. Nucleic Acids Res 45, D177-D182 (2017). https://doi.org/10.1093/nar/gkw1062 39 Aken, B. L. et al. Ensembl 2017. Nucleic Acids Res 45, D635-D642 (2017). https://doi.org/10.1093/nar/gkw1104 40 The UniProt Consortium. UniProt: the universal protein knowledgebase. Nucleic Acids Res 45, D158D169 (2017). https://doi.org/10.1093/nar/gkw1099 41 Rands, C. M., Meader, S., Ponting, C. P. & Lunter, G. 8.2% of the Human genome is constrained: variation in rates of turnover across functional element classes in the human lineage. PLoS Genet 10, e1004525 (2014). https://doi.org/10.1371/journal.pgen.1004525 42 Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. Nature 596, 583-589 (2021). https://doi.org/10.1038/s41586-021-03819-2 43 Baek, M. et al. Accurate prediction of protein structures and interactions using a three-track neural network. Science 373, 871-876 (2021). https://doi.org/10.1126/science.abj8754 44 Binder, J. L. et al. AlphaFold illuminates half of the dark human proteins. Curr Opin Struct Biol 74, 102372 (2022). https://doi.org/10.1016/j.sbi.2022.102372 45 Ashburner, M. et al. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25, 25-29 (2000). https://doi.org/10.1038/75556 46 Mi, H., Muruganujan, A., Casagrande, J. T. & Thomas, P. D. Large-scale gene function analysis with the PANTHER classification system. Nat Protoc 8, 1551-1566 (2013). https://doi.org/10.1038/nprot.2013.092 47 Sherman, B. T. et al. DAVID: a web server for functional enrichment analysis and functional annotation of gene lists (2021 update). Nucleic Acids Res 50, W216-W221 (2022). https://doi.org/10.1093/nar/gkac194 48 Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc Natl Acad Sci U S A 102, 15545-15550 (2005). https://doi.org/10.1073/pnas.0506580102 49 Hänzelmann, S., Castelo, R. & Guinney, J. GSVA: gene set variation analysis for microarray and RNAseq data. BMC Bioinformatics 14, 7 (2013). https://doi.org/10.1186/1471-2105-14-7 50 Chen, Y. et al. A Novel Immune-Related Gene Signature to Identify the Tumor Microenvironment and Prognose Disease Among Patients With Oral Squamous Cell Carcinoma Patients Using ssGSEA: A Bioinformatics and Biological Validation Study. Front Immunol 13, 922195 (2022). https://doi.org/10.3389/fimmu.2022.922195 25 | Page 51 Croft, D. et al. Reactome: a database of reactions, pathways and biological processes. Nucleic Acids Res 39, D691-697 (2011). https://doi.org/10.1093/nar/gkq1018 52 Kanehisa, M. & Goto, S. KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res 28, 2730 (2000). https://doi.org/10.1093/nar/28.1.27 53 Kuleshov, M. V. et al. Enrichr: a comprehensive gene set enrichment analysis web server 2016 update. Nucleic Acids Res 44, W90-97 (2016). https://doi.org/10.1093/nar/gkw377 54 Phan, A., Joshi, P., Kadelka, C. & Friedberg, I. A longitudinal analysis of function annotations of the human proteome reveals consistently high biases. Database (Oxford) 2025 (2025). https://doi.org/10.1093/database/baaf036 55 Timmons, J. A., Szkop, K. J. & Gallagher, I. J. Multiple sources of bias confound functional enrichment analysis of global -omics data. Genome Biol 16, 186 (2015). https://doi.org/10.1186/s13059-015-0761-7 56 Wijesooriya, K., Jadaan, S. A., Perera, K. L., Kaur, T. & Ziemann, M. Urgent need for consistent standards in functional enrichment analysis. PLoS Comput Biol 18, e1009935 (2022). https://doi.org/10.1371/journal.pcbi.1009935 57 Haynes, W. A., Tomczak, A. & Khatri, P. Gene annotation bias impedes biomedical research. Sci Rep 8, 1362 (2018). https://doi.org/10.1038/s41598-018-19333-x 58 Schnoes, A. M., Ream, D. C., Thorman, A. W., Babbitt, P. C. & Friedberg, I. Biases in the experimental annotations of protein function and their effect on our understanding of protein function space. PLoS Comput Biol 9, e1003063 (2013). https://doi.org/10.1371/journal.pcbi.1003063 59 Brunet, J. P., Tamayo, P., Golub, T. R. & Mesirov, J. P. Metagenes and molecular pattern discovery using matrix factorization. Proc Natl Acad Sci U S A 101, 4164-4169 (2004). https://doi.org/10.1073/pnas.0308531101 60 Devarajan, K. Nonnegative matrix factorization: an analytical and interpretive tool in computational biology. PLoS Comput Biol 4, e1000029 (2008). https://doi.org/10.1371/journal.pcbi.1000029 61 Berglund, A. E., Welsh, E. A. & Eschrich, S. A. Characteristics and Validation Techniques for PCABased Gene-Expression Signatures. Int J Genomics 2017, 2354564 (2017). https://doi.org/10.1155/2017/2354564 62 Alkaabi, A. M. & Abdallah, A. K. Portfolio practices in the principal evaluation process: A qualitative case study. Heliyon 10, e39467 (2024). https://doi.org/10.1016/j.heliyon.2024.e39467 63 Kong, W., Vanderburg, C. R., Gunshin, H., Rogers, J. T. & Huang, X. A review of independent component analysis application to microarray gene expression data. Biotechniques 45, 501-520 (2008). https://doi.org/10.2144/000112950 64 Argelaguet, R. et al. MOFA+: a statistical framework for comprehensive integration of multi-modal single-cell data. Genome Biol 21, 111 (2020). https://doi.org/10.1186/s13059-020-02015-1 65 Argelaguet, R. et al. Multi-Omics Factor Analysis-a framework for unsupervised integration of multiomics data sets. Mol Syst Biol 14, e8124 (2018). https://doi.org/10.15252/msb.20178124 66 Aibar, S. et al. SCENIC: single-cell regulatory network inference and clustering. Nat Methods 14, 10831086 (2017). https://doi.org/10.1038/nmeth.4463